<?xml version="1.0" encoding="UTF-8"?><rss xmlns:dc="http://purl.org/dc/elements/1.1/" xmlns:content="http://purl.org/rss/1.0/modules/content/" xmlns:atom="http://www.w3.org/2005/Atom" version="2.0" xmlns:cc="http://cyber.law.harvard.edu/rss/creativeCommonsRssModule.html">
    <channel>
        <title><![CDATA[Stories by Trivedi Aarsh on Medium]]></title>
        <description><![CDATA[Stories by Trivedi Aarsh on Medium]]></description>
        <link>https://medium.com/@trivediaarsh51?source=rss-607e7db4451b------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*dmbNkD5D-u45r44go_cf0g.png</url>
            <title>Stories by Trivedi Aarsh on Medium</title>
            <link>https://medium.com/@trivediaarsh51?source=rss-607e7db4451b------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 09:16:18 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@trivediaarsh51/feed" rel="self" type="application/rss+xml"/>
        <webMaster><![CDATA[yourfriends@medium.com]]></webMaster>
        <atom:link href="http://medium.superfeedr.com/" rel="hub"/>
        <item>
            <title><![CDATA[TryHackMe Walkthrough: CVE-2026–31431 — Copy Fail]]></title>
            <link>https://medium.com/@trivediaarsh51/tryhackme-walkthrough-cve-2026-31431-copy-fail-f85b9c58aa6b?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/f85b9c58aa6b</guid>
            <category><![CDATA[docker]]></category>
            <category><![CDATA[privilege-escalation]]></category>
            <category><![CDATA[linux]]></category>
            <category><![CDATA[tryhackme-walkthrough]]></category>
            <category><![CDATA[cve]]></category>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Sun, 15 Nov 2026 05:17:43 GMT</pubDate>
            <atom:updated>2026-11-15T05:17:43.955Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>Understanding and Exploiting a Linux Kernel Local Privilege Escalation Vulnerability</strong></p><p>Linux privilege escalation vulnerabilities are always interesting, especially when they allow an ordinary user to gain root access without requiring complicated techniques or unreliable timing.</p><p>In this TryHackMe room, we’ll explore <strong>CVE-2026–31431, also known as Copy Fail</strong>, a Linux kernel vulnerability that allows an unprivileged local user to modify the in-memory page cache of a file and use that capability to obtain root privileges.</p><p>What makes this vulnerability particularly interesting is that it doesn’t directly modify the file stored on disk. Instead, it tampers with the copy of the file held in memory.</p><h3>Task 1 — Introduction</h3><p>The first task introduces us to Copy Fail and explains why this vulnerability is worth understanding.</p><p>CVE-2026–31431 is a Linux kernel local privilege escalation vulnerability that takes advantage of an interaction between several kernel components.</p><p>The vulnerability allows an unprivileged user to perform controlled writes to the page cache of readable files. By modifying the cached contents of a privileged executable, an attacker can potentially execute malicious code with elevated privileges.</p><p>A particularly interesting aspect of Copy Fail is its reliability. Unlike many privilege escalation exploits that depend on race conditions or specific memory offsets, this vulnerability uses a deterministic write primitive.</p><p>The vulnerability was introduced by a kernel optimisation in 2017 and remained present for years before being discovered.</p><h3>Why is Copy Fail interesting?</h3><p>If you’ve previously studied <strong>Dirty Pipe (CVE-2022–0847)</strong>, some of the concepts here might sound familiar.</p><p>Both vulnerabilities involve modifying file contents through the Linux page cache without changing the actual file on disk.</p><p>However, they use different kernel mechanisms:</p><ul><li>Dirty Pipe abuses an issue involving pipe buffers.</li><li>Copy Fail takes advantage of the AF_ALG crypto interface and an AEAD implementation.</li></ul><p>The important takeaway is that modifying a file’s in-memory representation is not the same as modifying its on-disk contents.</p><h3>Learning objectives</h3><p>By completing this room, we’ll understand:</p><ul><li>How the Linux page cache works.</li><li>What AF_ALG does and why it matters.</li><li>How the authencesn algorithm contributes to the vulnerability.</li><li>Why splice() is important to the exploitation chain.</li><li>How Copy Fail can lead to local privilege escalation.</li><li>How defenders can detect and mitigate this vulnerability.</li></ul><p>The TryHackMe machine starts with an ordinary user account named karen, which we&#39;ll use throughout the practical section.</p><h3>Task 2 — Understanding the Vulnerability</h3><p>Before jumping into exploitation, it’s important to understand what actually makes Copy Fail possible.</p><p>The vulnerability comes from the interaction of four Linux kernel components:</p><ol><li>Page Cache</li><li>AF_ALG</li><li>authencesn</li><li>splice()</li></ol><p>Let’s understand each one.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*zCZMrlfznLSMklIS.png" /></figure><h3>1. Linux Page Cache</h3><p>The Linux page cache is an area of memory used by the kernel to store recently accessed file contents.</p><p>Whenever an application reads a file, Linux can load its contents into memory. Subsequent reads can then use the cached copy instead of repeatedly accessing the storage device.</p><p>This improves system performance considerably.</p><p>However, the page cache is shared between processes using the same kernel.</p><p>That means if its contents are modified, other processes accessing the same file may also see those modified contents.</p><p>This is the key concept behind Copy Fail.</p><p>The exploit changes the cached version of a file without modifying its original contents on disk.</p><p>As a result, traditional file integrity tools such as AIDE and Tripwire may not detect the modification because they inspect the actual file stored on disk.</p><h3>2. AF_ALG — Linux Crypto Socket Interface</h3><p>AF_ALG is a Linux socket interface that allows userspace applications to communicate with the kernel’s cryptographic subsystem.</p><p>It supports different cryptographic operations, including encryption, decryption, hashing, and authentication.</p><p>One important detail is that ordinary users can access AF_ALG without requiring root privileges.</p><p>Copy Fail takes advantage of this interface to reach the vulnerable cryptographic implementation.</p><h3>3. authencesn — The Important AEAD Component</h3><p>The next component is authencesn.</p><p>It is an AEAD (Authenticated Encryption with Associated Data) template used by the Linux kernel, including in IPsec-related cryptographic operations.</p><p>During a decryption operation, authencesn performs a four-byte scratch write involving the extended sequence number.</p><p>The problem is related to the order in which the operations happen.</p><p>The scratch write occurs before authentication verification is completed.</p><p>Even if the subsequent HMAC verification fails, the earlier write is not automatically reversed.</p><p>This behaviour becomes dangerous when the output buffer points to memory belonging to a file’s page cache.</p><h3>4. splice() — Moving Page References</h3><p>The splice() system call allows data to be transferred between file descriptors without the usual userspace copying operation.</p><p>Instead of creating an ordinary copy of the file contents, the kernel can transfer references to the underlying memory pages.</p><p>This is normally a performance optimisation.</p><p>However, Copy Fail abuses this behaviour to make the AF_ALG processing pipeline operate on pages belonging to a regular file’s page cache.</p><h3>The 2017 Kernel Optimisation</h3><p>Now we get to the root cause.</p><p>In 2017, a kernel optimisation changed how the AF_ALG AEAD implementation handled source and destination scatterlists.</p><p>The optimisation introduced an in-place processing approach in which the source and destination references could point to the same scatterlist.</p><p>This worked for the normal data path, where the buffers were supplied through ordinary userspace writes.</p><p>However, the implementation did not properly account for data supplied through splice().</p><p>When file-backed page cache pages entered this processing path, the output operation could end up modifying those same pages.</p><p>Combined with the scratch write performed by authencesn, this created the vulnerability.</p><h3>Putting Everything Together</h3><p>The exploitation chain can be understood as follows:</p><ol><li>An unprivileged user opens an AF_ALG socket.</li><li>The socket is configured to use the vulnerable AEAD implementation.</li><li>File-backed pages are passed into the socket through splice().</li><li>The in-place processing behaviour causes the output buffer to reference the file’s page cache.</li><li>The authencesn scratch operation performs a controlled four-byte write.</li><li>Authentication subsequently fails, but the write remains.</li><li>Repeating the operation allows multiple chunks to be modified.</li></ol><p>The result is a controlled page-cache write primitive.</p><p>This is what makes Copy Fail particularly dangerous.</p><h3>Task 2 Questions and Answers</h3><p><strong>Q1. What year was the optimisation introduced that created this vulnerability?</strong></p><p>Answer: 2017</p><p><strong>Q2. Which AEAD algorithm template performs the scratch write that corrupts the page cache?</strong></p><p>Answer: authencesn</p><p><strong>Q3. What system call transfers page cache pages into the AF_ALG socket without copying them?</strong></p><p>Answer: splice</p><p><strong>Q4. The HMAC verification fails after the scratch write is performed. Does this undo the page cache corruption?</strong></p><p>Answer: Nay</p><h3>Task 3 — Exploitation</h3><p>Now comes the practical part!</p><p>In this task, we’ll use the vulnerable TryHackMe machine to understand how Copy Fail can be used to obtain root privileges.</p><p>The machine starts with an unprivileged user named karen.</p><p>The provided proof of concept is located at:</p><p>/home/karen/exploit.py</p><p>The PoC targets /usr/bin/su, a privileged executable that can be used to obtain elevated privileges.</p><h3>Step 1 — Check the Current User</h3><p>First, let’s confirm which user we’re logged in as.</p><pre>id</pre><p>The output should look similar to this:</p><pre>uid=1001(karen) gid=1001(karen) groups=1001(karen)</pre><p>As expected, we’re operating as a normal user without root privileges.</p><p>This is important because the vulnerability is designed to demonstrate local privilege escalation from an unprivileged account.</p><h3>Step 2 — Inspect the Exploit</h3><p>Before running the provided PoC, let’s take a quick look at the script.</p><pre>head -30 /home/karen/exploit.py</pre><p>The script uses Python’s standard library and interacts with Linux kernel interfaces.</p><p>Its main components include:</p><ul><li>socket for AF_ALG communication.</li><li>os for system-level operations.</li><li>zlib for CRC-related calculations.</li></ul><p>The overall process uses the vulnerable crypto interface and page-cache behaviour to modify the target executable in memory.</p><p><strong>Note:</strong> Understanding what a proof of concept does before executing it is always a good habit, particularly when working with kernel-level vulnerabilities.</p><h3>Step 3 — Execute the Proof of Concept</h3><p>With the initial checks completed, the next step is to execute the provided PoC in the TryHackMe lab.</p><p>The exploit demonstrates how the vulnerable kernel behaviour can be used to modify the cached contents of the target executable and obtain elevated privileges.</p><p>Once execution succeeds, the shell should provide root privileges.</p><h3>Step 4 — Verify Root Access</h3><p>Let’s check our current identity:</p><pre>whoami</pre><p>Expected output:</p><pre>root</pre><p>We can also verify the effective user ID:</p><pre>id</pre><p>The output should indicate:</p><pre>uid=0(root) gid=0(root) groups=0(root)</pre><p>And just like that, we’ve demonstrated local privilege escalation!</p><p>We started as an ordinary user named karen and successfully obtained root access.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*iFLNaWxAlTcY8Tzd.png" /></figure><h3>Step 5 — Retrieve the Flag</h3><p>Now that we have root access, we can access the root user’s flag file.</p><pre>cat /root/flag.txt</pre><p><strong>Flag:</strong></p><p>THM{copy_fail_kernel_lpe}</p><h3>Step 6 — The Interesting Part: Checking the File Hash</h3><p>Here’s one of the most interesting demonstrations in this room.</p><p>Even after successfully exploiting /usr/bin/su, the actual file stored on disk remains unchanged.</p><p>Let’s verify that using:</p><pre>sha256sum /usr/bin/su</pre><p>The expected output in the lab is:</p><pre>c4d2e053445c5f89d13b68bb54de8d67358e1aa20a2b8f0688cb8a47a32edbdf  /usr/bin/su</pre><p>The hash remains consistent with the original file.</p><p>Why?</p><p>Because the exploit modifies the in-memory page cache rather than writing the changes to the actual file on disk.</p><p>This means traditional file integrity checks can report that the executable is unchanged, even though its cached representation was temporarily modified.</p><p>That’s a significant security concern.</p><h3>Step 7 — Exit the Root Shell</h3><p>Once you’ve retrieved the flag and completed the verification, exit the elevated shell.</p><pre>exit</pre><figure><img alt="" src="https://cdn-images-1.medium.com/max/875/0*bvK4Vuhyfh5mJy3Z.png" /></figure><p>The PoC also performs cleanup by advising the kernel that the modified cached pages are no longer needed.</p><p>This allows the kernel to evict those pages and reload the original file contents from disk when required.</p><p>This behaviour also explains why detecting the vulnerability after exploitation can be difficult.</p><p>The evidence may disappear quickly.</p><h3>Task 4 — Detection and Remediation</h3><p>Obtaining root access is only one side of the story.</p><p>From a defensive perspective, the next question is:</p><p><strong>How can we detect and prevent Copy Fail?</strong></p><p>As we’ve already seen, checking file hashes is not sufficient because the exploit does not need to modify the on-disk file.</p><p>Instead, defenders need to focus on suspicious process behaviour and kernel interactions.</p><h3>Detection — Monitoring AF_ALG Socket Creation</h3><p>One of the most important indicators is the creation of AF_ALG sockets.</p><p>The exploit repeatedly interacts with the AF_ALG interface to perform its page-cache modifications.</p><p>A particularly useful detection detail is the socket type.</p><p>AF_ALG sockets using SOCK_SEQPACKET are associated with AEAD operations and can provide a more specific detection signal than monitoring AF_ALG usage alone.</p><p>The socket address family number for AF_ALG is:</p><pre>38</pre><p>Linux audit rules can monitor socket creation using this value.</p><p>The room provides the following example:</p><pre>-a always,exit -F arch=b64 -S socket -F a0=38 -k copy_fail_af_alg<br>-a always,exit -F arch=b64 -S splice -k copy_fail_splice</pre><p>These rules help identify processes interacting with the relevant kernel interfaces.</p><h3>Monitoring Suspicious Processes</h3><p>Not every AF_ALG socket is malicious.</p><p>Legitimate applications also use the interface for cryptographic operations.</p><p>Examples include:</p><ul><li>cryptsetup</li><li>systemd-cryptsetup</li><li>kcapi-enc</li><li>kcapi-dgst</li><li>kcapi-mac</li><li>kcapi-speed</li><li>strongSwan components</li></ul><p>Therefore, simply detecting an AF_ALG socket should not automatically be treated as proof of exploitation.</p><p>A better approach is to correlate multiple events.</p><p>For example:</p><ul><li>Unexpected AF_ALG socket creation.</li><li>Repeated socket operations within a short period.</li><li>Unusual splice() activity.</li><li>Repeated recvmsg() calls.</li><li>Execution of a privileged binary immediately afterwards.</li></ul><p>Combining these signals can provide much stronger evidence of a potential Copy Fail exploitation attempt.</p><h3>Falco Detection</h3><p>The room also introduces Falco as an example of a runtime security monitoring solution.</p><p>Falco can monitor system activity and generate alerts when suspicious behaviour occurs.</p><p>A detection rule can identify AF_ALG socket creation by processes that are not part of an approved application list.</p><p>The important idea is to establish a baseline of legitimate AF_ALG usage and investigate unexpected activity.</p><p>This is especially useful in environments running multiple services or containers.</p><h3>Mitigation — Disabling algif_aead</h3><p>One temporary mitigation discussed in the room is disabling the vulnerable algif_aead kernel module.</p><p>Before applying this approach, we need to determine whether the component is available as a loadable module.</p><p>The command used in the room is:</p><pre>modinfo algif_aead</pre><p>If module information is returned, the module-based mitigation may be applicable.</p><h3>Applying the Modprobe Configuration</h3><p>On systems where algif_aead is a loadable module, the room demonstrates the following configuration:</p><pre>echo &quot;install algif_aead /bin/false&quot; | sudo tee /etc/modprobe.d/disable-algif-aead.conf</pre><p>This configuration instructs modprobe to execute /bin/false instead of loading the module.</p><p>If the module is currently loaded, it can also be unloaded.</p><p>After applying the configuration, attempting to load the module should fail.</p><p>This prevents the vulnerable module from being loaded through the normal modprobe mechanism.</p><p><strong>Important:</strong> This is an interim mitigation, not a replacement for installing the patched kernel.</p><h3>What About RHEL-Based Distributions?</h3><p>The room also highlights an important difference between Linux distributions.</p><p>On some RHEL-family systems, the relevant functionality may be compiled directly into the kernel instead of being provided as a loadable module.</p><p>In that situation, a modprobe blacklist will not provide the intended protection.</p><p>The correct mitigation depends on how the kernel was built and how the distribution packages its kernel components.</p><p>Always follow the security guidance provided by your Linux distribution.</p><h3>Permanent Fix — Update the Kernel</h3><p>The recommended long-term solution is to install a kernel containing the official security fix.</p><p>The room identifies the fixed mainline versions as:</p><ul><li>Linux 6.18.22</li><li>Linux 6.19.12</li><li>Linux 7.0</li></ul><p>Distribution-specific backported fixes may also be available, so version numbers alone should not be used to determine whether a particular vendor kernel is vulnerable.</p><p>Check your distribution’s security advisories and install the appropriate patched kernel.</p><h3>Task 4 Question and Answer</h3><p><strong>Q1. What command checks whether algif_aead is a loadable module or compiled into the kernel?</strong></p><p>Answer:</p><p>modinfo algif_aead</p><h3>Task 5 — Conclusion</h3><p>And that’s the end of the Copy Fail room!</p><p>This room was a great demonstration of how seemingly harmless kernel optimisations can introduce serious security vulnerabilities.</p><p>Here are a few important things I learned:</p><p><strong>1. The page cache is an important attack surface.</strong></p><p>The page cache is shared kernel infrastructure. Modifying its contents can affect how other processes access files, even when the original files on disk remain untouched.</p><p><strong>2. File integrity monitoring has limitations.</strong></p><p>Tools that only calculate hashes of files stored on disk may not detect memory-level modifications.</p><p><strong>3. Kernel interfaces deserve careful security review.</strong></p><p>Copy Fail demonstrates how components that work correctly individually can become vulnerable when combined in an unexpected way.</p><p><strong>4. Behaviour-based detection matters.</strong></p><p>Monitoring system calls, unusual AF_ALG activity, and suspicious process behaviour can help identify exploitation attempts that traditional filesystem monitoring misses.</p><p><strong>5. Keeping the kernel updated is essential.</strong></p><p>Temporary mitigations can reduce exposure, but installing the appropriate security patches is the best long-term solution.</p><h3>Final Thoughts</h3><p>I really enjoyed working through this TryHackMe room because it goes beyond simply running an exploit and retrieving a flag.</p><p>It explains the underlying Linux kernel behaviour and demonstrates why an attacker doesn’t always need to modify a file on disk to compromise a system.</p><p>The most interesting part for me was seeing how a page-cache modification could affect executable behaviour while leaving the original file hash unchanged.</p><p>It was also a good reminder that security monitoring should not rely entirely on filesystem integrity checks.</p><p><strong>Room completed!</strong></p><p>Thanks for reading, and I hope this walkthrough helped you understand CVE-2026–31431 and the Copy Fail vulnerability a little better.</p><p>If you enjoyed this write-up, feel free to clap, share your thoughts, and follow for more TryHackMe walkthroughs and cybersecurity content.</p><p>Happy hacking! 🚩</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=f85b9c58aa6b" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[TryHackMe Carnage Writeup]]></title>
            <link>https://medium.com/@trivediaarsh51/tryhackme-carnage-writeup-62c762a58746?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/62c762a58746</guid>
            <category><![CDATA[network-security]]></category>
            <category><![CDATA[pcap-analysis]]></category>
            <category><![CDATA[wireshark]]></category>
            <category><![CDATA[blue-teaming]]></category>
            <category><![CDATA[network-forensics]]></category>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Fri, 13 Nov 2026 12:07:27 GMT</pubDate>
            <atom:updated>2026-11-13T12:07:27.428Z</atom:updated>
            <content:encoded><![CDATA[<h4>Investigating a Malware Infection Through Network Forensics Using Wireshark</h4><p>Platform: TryHackMe<br>Room: Carnage<br>Category: Blue Team | Network Forensics | Incident Response<br>Tool: Wireshark<br>Difficulty: Easy<br>Investigation Date: September 24, 2021 (PCAP timeline)</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1000/0*aILyABBsMp48wr4T.png" /></figure><h3>1. Introduction</h3><p>What if a seemingly harmless Excel spreadsheet was the beginning of a complete malware infection?</p><p>As Blue Team analysts, one of our most important responsibilities is identifying suspicious network activity, understanding how attackers communicate with compromised systems, and collecting evidence to reconstruct security incidents.</p><p>In this blog, I will walk through my investigation of the <strong>Carnage room on TryHackMe</strong>, where the objective was to analyse malicious network traffic using Wireshark.</p><p>The investigation involved examining HTTP traffic, identifying suspicious domains, analysing downloaded files, investigating Cobalt Strike Command and Control (C2) infrastructure, and identifying possible post-infection and email-related activity.</p><p>Instead of simply looking for answers, I approached this challenge from the perspective of a SOC analyst investigating a potentially compromised Windows machine.</p><p>Let’s get started!</p><h3>2. Investigation Environment</h3><p>For this investigation, I used the following tools and resources:</p><ul><li>TryHackMe — Carnage room</li><li>Wireshark — Network protocol analyser</li><li>PCAP file — Captured network traffic</li><li>VirusTotal — Threat intelligence and IP reputation analysis</li></ul><p>The victim machine identified in the investigation was:</p><p>10.9.23.102</p><p>The objective was to analyse the provided PCAP file and identify suspicious communications, malicious domains, C2 infrastructure, and other indicators of compromise.</p><h3>3. Initial HTTP Traffic Investigation</h3><p>The first step was to identify when the victim machine initially communicated with the malicious infrastructure.</p><p>I opened the provided PCAP file in Wireshark and applied the following display filter:</p><pre>http</pre><p>This helped narrow down the captured traffic to HTTP communications.</p><p>By examining the Frame section and the packet arrival timestamp, I identified the first HTTP connection to the malicious IP.</p><p><strong>Finding:</strong></p><p>2021-09-24 16:44:38</p><p>This timestamp provided the starting point for reconstructing the incident timeline.</p><h3>Identifying the Downloaded ZIP File</h3><p>While examining the HTTP traffic, I identified a ZIP file downloaded by the victim machine.</p><p>The filename was:</p><p>documents.zip</p><p>This was an important finding because it provided a possible entry point for the malware infection.</p><p>Rather than downloading or executing the suspicious file, I focused on examining the available network evidence.</p><h3>Identifying the Malicious Hosting Domain</h3><p>Next, I investigated the IP address associated with the suspicious HTTP communication.</p><p>I applied the following Wireshark display filter:</p><pre>ip.addr == 85.187.128.24</pre><p>By examining the HTTP packet details, I identified the domain associated with the malicious file download.</p><p><strong>Finding:</strong></p><p>attirenepal.com</p><p>An IP address alone does not always provide enough context during an investigation. Identifying the associated domain helps analysts understand the infrastructure involved in suspicious activity.</p><figure><img alt="Initial HTTP Traffic Investigation" src="https://cdn-images-1.medium.com/max/1024/1*gmcVwjyxsC15pNdGCaYzcw.png" /><figcaption>Initial HTTP Traffic Investigation</figcaption></figure><figure><img alt="Identifying the Malicious Hosting Domain" src="https://cdn-images-1.medium.com/max/1024/1*Df_njiznVPoEY1ANwb73bQ.png" /><figcaption>Identifying the Malicious Hosting Domain</figcaption></figure><h3>4. Identifying the File Inside the ZIP Archive</h3><p>The next question was particularly interesting.</p><p>Without downloading the file, I needed to identify the filename contained inside the ZIP archive.</p><p>I used Wireshark’s <strong>Follow HTTP Stream</strong> functionality to examine the relevant HTTP communication.</p><p>The filename identified was:</p><p>chart-1530076591.xls</p><p>This indicated that the downloaded archive contained an Excel spreadsheet.</p><p>Spreadsheets are commonly used in phishing and malware delivery campaigns. They can potentially contain malicious macros, embedded content, or exploit attempts.</p><p>However, the filename alone does not establish how the file executed or whether it contained a particular payload.</p><p>This is why correlating network evidence with endpoint telemetry is important.</p><h3>5. Analysing the Web Server Information</h3><p>I continued examining the same HTTP stream to identify information about the web server hosting the malicious ZIP file.</p><p>The HTTP response revealed the following information:</p><p><strong>Web Server:</strong></p><p>LiteSpeed</p><p><strong>Server Information:</strong></p><p>PHP/7.2.34</p><p>These details provided additional context about the infrastructure involved in delivering the suspicious archive.</p><p>One important distinction is that PHP and LiteSpeed are different technologies. The PHP version should not be mistaken for the LiteSpeed web server version.</p><p>This information is useful when documenting infrastructure observed during a network investigation.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3q4dXSNPGK8rtJYkSNQb8Q.png" /></figure><h3>6. Identifying Additional Malicious Domains</h3><p>The investigation then moved towards identifying additional domains involved in downloading potentially malicious files to the victim’s computer.</p><p>I used the following Wireshark display filter:</p><pre>tls.handshake.type == 1 &amp;&amp; (frame.time &gt;= &quot;2021-09-24 16:45:11&quot; &amp;&amp; frame.time &lt;= &quot;2021-09-24 16:45:30&quot;)</pre><p>This filter helped narrow down TLS Client Hello packets within the relevant time window.</p><p>By examining the handshake information and associated domain names, I identified three domains involved in this activity:</p><ul><li>finejewels.com.au</li><li>thietbiagt.com</li><li>new.americold.com</li></ul><p>These findings expanded the investigation beyond the initial ZIP download.</p><p>TLS handshake metadata, including Server Name Indication (SNI) when available, can help analysts identify destination domains even when the application traffic itself is encrypted.</p><p>This is particularly useful during network threat hunting.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*d-OpGr_1mtLFaB72CJzhww.png" /><figcaption>Identifying Additional Malicious Domains</figcaption></figure><h3>7. Investigating the SSL Certificate Authority</h3><p>Next, I focused on the first domain:</p><p>finejewels.com.au</p><p>By examining the certificate information associated with its TLS communication, I identified the Certificate Authority that issued the SSL certificate.</p><p><strong>Finding:</strong></p><p>GoDaddy</p><p>This was an interesting observation because it demonstrated that malicious or compromised infrastructure can still use certificates issued by recognised Certificate Authorities.</p><p>A valid SSL certificate does not automatically mean that a website is safe.</p><p>As security analysts, we should never rely on HTTPS or certificate validity alone when determining whether a domain is trustworthy.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*ONIksJbFRRM-NRVNXD2Bpw.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/853/1*PMAdZA9o0qwajyFf4U3VUA.png" /></figure><h3>8. Identifying Cobalt Strike Command and Control Infrastructure</h3><p>One of the most interesting parts of this investigation was identifying potential Cobalt Strike Command and Control infrastructure.</p><p>Cobalt Strike is a commercial adversary simulation platform that is also frequently abused by threat actors.</p><p>Its Beacon component can communicate with remote Command and Control servers, allowing operators to interact with compromised systems.</p><p>To investigate suspicious external connections, I started by filtering common web traffic ports.</p><p>The Wireshark display filter was:</p><pre>tcp.port == 80 or tcp.port == 8080</pre><p>Next, I opened:</p><p><strong>Statistics → Conversations</strong></p><p>I examined the IPv4 and TCP conversation statistics, paying particular attention to external connections communicating with the victim host:</p><p>10.9.23.102</p><p>I sorted the conversations by packet count to identify connections worth investigating.</p><p>The suspicious IP addresses were then checked using VirusTotal, particularly its Community tab.</p><p>The investigation identified two IP addresses associated with Cobalt Strike C2 infrastructure:</p><ul><li>185.106.96.158</li><li>185.125.204.174</li></ul><p>This was an important finding because it suggested that the victim machine was communicating with infrastructure associated with command-and-control activity.</p><p>Combining network traffic analysis with threat intelligence helped provide stronger context for these suspicious connections.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*o16KFxvssV8rR5StNd9Uqg.png" /></figure><h3>9. Investigating the Cobalt Strike Domains</h3><p>After identifying the two suspicious IP addresses, I continued investigating their associated HTTP communications and domain information.</p><h3>First Cobalt Strike IP</h3><p>IP address:</p><p>185.106.96.158</p><p>By examining the HTTP GET request and its Host header, I identified the following value:</p><p>oscp.verisign.com</p><p>The associated C2 domain identified during the investigation was:</p><p>survmeter.live</p><h3>Second Cobalt Strike IP</h3><p>IP address:</p><p>185.125.204.174</p><p>The associated C2 domain identified was:</p><p>securitybusinpuff.com</p><p>These findings helped establish a connection between the suspicious IP addresses and their associated domain infrastructure.</p><p>It is important to distinguish between an HTTP Host header and a domain associated with an IP through historical DNS records or threat intelligence. They are not necessarily the same thing.</p><p>This distinction is important when documenting indicators of compromise because inaccurate domain-to-IP relationships can lead to incorrect conclusions.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*q1a4FpHQQvj0-PEQlfqQzg.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/932/1*caWy2dYuuVV_ESWhzzmlzA.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*WBV-GHnpvIe21g0QY9YTwQ.png" /></figure><h3>10. Investigating Post-Infection Traffic</h3><p>After identifying the suspicious C2 infrastructure, I moved towards examining post-infection network communications.</p><p>I focused on HTTP POST requests because these can contain information transmitted from a compromised system to a remote server.</p><p>By filtering the traffic for HTTP POST requests and examining the Host header, I identified the domain:</p><p>maldivehost.net</p><p>This domain was associated with post-infection communication in the captured traffic.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/937/1*AdgXP6W7JnDZrhACww2XTQ.png" /></figure><h3>Identifying the First Eleven Characters</h3><p>The next step was to inspect the URI path of the first outgoing HTTP POST request.</p><p>The first eleven characters sent by the victim host were:</p><p>zLIisQRWZI9</p><p>These characters were identified from the URI path:</p><p>/zLIisQRWZI9/...</p><p>This was an interesting observation because HTTP request paths can contain identifiers or application-specific information.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/910/1*x_F7WJVna159XsIbobfAWQ.png" /></figure><h3>Identifying the First Packet Length</h3><p>The length of the first packet sent to the C2 server was:</p><p>281 bytes</p><p>This value was identified using the Length column in Wireshark’s packet list.</p><h3>Identifying the Server Header</h3><p>I then examined the relevant HTTP stream and inspected the server response headers.</p><p>The Server header was:</p><p>Apache/2.4.49 (cPanel) OpenSSL/1.1.1l mod_bwlimited/1.4</p><p>This provided additional information about the server handling the suspicious communication.</p><p>Together, these observations helped document the post-infection network activity.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*IRIPS16KkxeFMCJslaEdEQ.png" /></figure><h3>11. Investigating the IP Address Lookup Activity</h3><p>Another interesting observation was that the malware-related activity involved checking the victim machine’s public IP address.</p><p>During the investigation, I examined DNS traffic and identified a query associated with:</p><p>api.ipify.org</p><p>This domain is associated with a service that returns the public IP address of a client.</p><p>The DNS query occurred at:</p><p>2021-09-24 17:00:04 UTC</p><p>This finding is relevant because malware and remote-access tools may obtain the public IP address of a compromised machine to understand its external network environment.</p><p><strong>Findings:</strong></p><ul><li>DNS domain: api.ipify.org</li><li>Query timestamp: 2021-09-24 17:00:04 UTC</li></ul><p>This is another example of why DNS analysis is an important part of network forensics.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*6jGXvHRB-CbDdKf2IXaltA.png" /></figure><h3>12. Investigating Malspam Activity</h3><p>The final part of the investigation involved examining SMTP traffic.</p><p>SMTP is commonly used for email transmission, making it particularly relevant when investigating phishing and spam-related activity.</p><p>I filtered the captured traffic using:</p><pre>smtp</pre><p>By examining the SMTP packets, I identified the first MAIL FROM address observed in the traffic.</p><p><strong>Finding:</strong></p><p>farshin@mailfa.com</p><p>I also examined the total number of packets associated with SMTP communication.</p><p><strong>Total SMTP packets:</strong></p><p>1439</p><p>These findings added another layer to the investigation by showing that the captured traffic included email-related activity.</p><p>Email-related indicators are particularly useful when correlating network investigations with phishing investigations and mail server logs.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*dIRig_j02KfjF2YTekHURg.png" /></figure><h3>13. Key Indicators of Compromise</h3><p>During this investigation, I identified several indicators that could be useful for documenting and correlating suspicious activity.</p><p><strong>Initial malicious download infrastructure:</strong></p><ul><li>IP address: 85.187.128.24</li><li>Domain: attirenepal.com</li><li>Downloaded archive: documents.zip</li><li>File inside archive: chart-1530076591.xls</li></ul><p><strong>Additional suspicious domains:</strong></p><ul><li>finejewels.com.au</li><li>thietbiagt.com</li><li>new.americold.com</li></ul><p><strong>Cobalt Strike-related infrastructure:</strong></p><ul><li>185.106.96.158</li><li>185.125.204.174</li><li>survmeter.live</li><li>securitybusinpuff.com</li></ul><p><strong>Post-infection communication:</strong></p><ul><li>Domain: maldivehost.net</li><li>URI identifier: zLIisQRWZI9</li></ul><p><strong>Other observations:</strong></p><ul><li>IP lookup service: api.ipify.org</li><li>First SMTP sender: farshin@mailfa.com</li></ul><p>These indicators should be treated as historical observations from the supplied PCAP and investigated in context before being used in production detection rules or blocklists.</p><h3>14. What Did I Learn as a Blue Team Analyst?</h3><p>This investigation helped me understand how much information can be extracted from a PCAP file.</p><p>Here are some of my key takeaways.</p><p><strong>1. HTTP traffic can reveal suspicious downloads.</strong></p><p>Examining HTTP requests and responses can help identify suspicious files, domains, and server information.</p><p><strong>2. TLS handshakes provide valuable investigation clues.</strong></p><p>Even when application traffic is encrypted, metadata such as SNI, when available, can help identify destination domains.</p><p><strong>3. C2 infrastructure identification requires correlation.</strong></p><p>A suspicious IP address alone is not sufficient evidence. Combining packet analysis with threat intelligence helps build a stronger investigation.</p><p><strong>4. HTTP POST requests deserve attention.</strong></p><p>POST traffic can be relevant when investigating possible data transmission or malware communications.</p><p><strong>5. DNS analysis is essential.</strong></p><p>DNS queries can reveal which domains a compromised machine attempted to contact and help analysts identify suspicious external dependencies.</p><p><strong>6. Network forensics is about connecting evidence.</strong></p><p>The real value of an investigation comes from correlating timestamps, IP addresses, domains, protocols, and network conversations to build an incident timeline.</p><h3>15. Final Thoughts</h3><p>The Carnage room was a valuable hands-on exercise in network forensics and Blue Team investigation.</p><p>What initially looked like a suspicious ZIP download turned into a much broader investigation involving multiple domains, Cobalt Strike infrastructure, post-infection communications, public IP lookup activity, and SMTP traffic.</p><p>For anyone learning SOC analysis, I strongly recommend spending time understanding Wireshark rather than simply memorising display filters.</p><p>Knowing what to look for, why a packet is suspicious, and how different pieces of evidence connect is what makes a good security analyst.</p><p>This challenge helped me improve my understanding of network traffic analysis and gave me practical exposure to investigating suspicious communications from a Blue Team perspective.</p><p>I am continuing my journey towards becoming a better Blue Team and SOC analyst, one investigation at a time.</p><p><strong>TryHackMe Room:</strong> Carnage<br><a href="https://tryhackme.com/room/c2carnage">https://tryhackme.com/room/c2carnage</a></p><p>If you are also learning Blue Team operations, I encourage you to solve this room yourself and compare your findings.</p><p>Happy Hunting!</p><p>#Cybersecurity #BlueTeam #TryHackMe #Wireshark #NetworkForensics #SOCAnalyst #IncidentResponse #ThreatHunting #CobaltStrike #DFIR</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=62c762a58746" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[You’re in the Cave: My TryHackMe Journey Through XXE, Java Deserialization & Docker Escape |…]]></title>
            <link>https://medium.com/@trivediaarsh51/lost-in-the-cave-my-tryhackme-journey-through-xxe-java-deserialization-docker-escape-b43fb30dd553?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/b43fb30dd553</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Thu, 12 Nov 2026 10:04:38 GMT</pubDate>
            <atom:updated>2026-11-12T10:11:10.547Z</atom:updated>
            <content:encoded><![CDATA[<h3>You’re in the Cave: My TryHackMe Journey Through XXE, Java Deserialization &amp; Docker Escape | TryHackMe | Writeup</h3><h3>Introduction</h3><p>Some TryHackMe rooms are simple: scan the machine, discover a vulnerability, get a shell, escalate privileges, and capture the flags.</p><p>But occasionally, you encounter a room that makes you think differently.</p><p><strong>You’re in a Cave</strong> was one of that rooms for me.</p><p>What initially appeared to be a simple text-based room turned into an exciting journey involving web enumeration, XML External Entity (XXE), Java deserialization-Linux privilege escalation, and Docker container escape.</p><h3>1. Initial Reconnaissance</h3><p>As always, I started with reconnaissance.</p><p>Before interacting with the target, I wanted to understand which services were running and what ports were exposed.</p><p>The target machine was:</p><p><strong>Target IP:</strong> 10.48.133.224</p><p>I started with a full-port Nmap scan:</p><pre>nmap -sC -sV -p- 10.48.133.224</pre><p>The scan revealed three interesting ports:</p><ul><li><strong>Port 80:</strong> HTTP — Apache web server.</li><li><strong>Port 2222:</strong> SSH — OpenSSH service.</li><li><strong>Port 3333:</strong> A custom service !</li></ul><figure><img alt="My initial Nmap scan revealed the exposed services running on the target machine." src="https://cdn-images-1.medium.com/max/813/1*m0o7W-TAa9UYhG1abtp6Ww.png" /><figcaption>My initial Nmap scan revealed the exposed services running on the target machine.</figcaption></figure><p>Port 80 was hosting a web app, port 2222 was running on SSH &amp; port 3333 appeared to be something different.</p><p>When I connected to port 3333; ireceived a message asking what I wanted to do inside the cave.</p><p>It looked like to be a text-based adventure game.</p><p>I tried with few commands but I didn’t understand how the application worked.</p><p>So i moved on to the web server.</p><h3>2. Exploring the Web Application</h3><p>I started to check the target in my browser:</p><pre>http://10.48.133.224</pre><p>The page displayed a simple interface with an input field where I could interact with the application.</p><p>I tried few input but nothing particularly useful appeared.</p><p>My next step was directory enumeration.</p><p>I used Gobuster:</p><pre>gobuster dir -u http://10.48.133.224/ -w &lt;WORDLIST&gt;</pre><p>The enumeration revealed some interesting endpoints, including:</p><ul><li>/matches</li><li>/search</li></ul><figure><img alt="Directory enumeration helped me discover additional endpoints that were not immediately visible on the homepage." src="https://cdn-images-1.medium.com/max/807/1*oLgusnD6N9UQgUuT3o53bw.png" /><figcaption><em>Directory enumeration helped me discover additional endpoints that were not immediately visible on the homepage.</em></figcaption></figure><p>When I explored these endpoint i noticed that the responses carried Base64-encoded information.</p><p>I decoded the returned data &amp; inspect its contents.</p><p>The decoded information showed serialized Java objects.</p><p>CLUE !</p><p>Why does a simple adventure game returning serialized Java data?</p><p>I thought that the custom service on port 3333 and the web application were connected anyhow .</p><h3>3. Exploring the Cave</h3><p>I returned to the custom service and continued checking with the available commands.</p><p>One interaction revealed a matche to identify the current working directory:</p><pre>/home/cave/src</pre><p>This is interesting because the application is revealing information through the game interface.</p><p>I tried directory enumeration.</p><p>Some of the interesting endpoints included:</p><ul><li>/attack</li><li>/lamp</li><li>/walk</li></ul><p>After executing it, I discovered references to several application files:</p><pre>Action.class<br>RPG.class<br>RPG.java<br>Serialize.class<br>commons-io-2.7.jar<br>run.sh</pre><figure><img alt="The lamp interaction exposed useful information about the Java application." src="https://cdn-images-1.medium.com/max/687/1*Amx0j458bpH5Of9IXZslRw.png" /><figcaption><em>The lamp interaction exposed useful information about the Java application.</em></figcaption></figure><p>Finding RPG.java was good.</p><p>Instead of guessing how the application worked, I now can understand its internal logic.</p><p>But first, I needed to investigate how the web application processed user input.</p><h3>4. Discovering the XXE Vulnerability</h3><p>Now that I knew the location and filename of the Java source code, I started thinking about how to retrieve it.</p><p>I went back to the web application and focused on action.php.</p><p>While investigating the application, I discovered that it processed XML input.</p><p>This immediately made me think about <strong>XML External Entity (XXE) injection</strong>.</p><h3>What is XXE?</h3><p>XXE is a vulnerability that occurs when an application processes XML input using an improperly configured XML parser.</p><p>If external entities are enabled, an attacker may be able to make the application access local files or other resources that should not be accessible.</p><p>In this room, XXE became the key to retrieving the Java source code.</p><p>The hint provided by the room was particularly interesting:</p><p><em>Sometimes things that work with POST can also work with GET.</em></p><p>This gave me an idea.</p><p>Instead of relying on a traditional POST request, I investigated how the application handled XML data supplied through a GET request.</p><p>The application reflected the supplied content in its response.</p><p>This was exactly what I needed.</p><p>I used the XXE vulnerability to retrieve the contents of:</p><pre>/home/cave/src/RPG.java</pre><p>Finally, I had access to the source code!</p><figure><img alt="Using the XML processing vulnerability in action.php, I was able to retrieve the Java source code." src="https://cdn-images-1.medium.com/max/717/1*JFufUaSfqa9DEXS5Bq_e4Q.png" /><figcaption><em>Using the XML processing vulnerability in action.php, I was able to retrieve the Java source code.</em></figcaption></figure><h3>5. Understanding RPG.java</h3><p>After retrieving RPG.java, I spent some time understanding how the application worked.</p><p>The source code revealed that the custom service was listening on port 3333.</p><p>The most interesting part was how it processed the input received from the user.</p><p>The application constructed a URL using the input:</p><pre>URL url = new URL(&quot;http://cave.thm/&quot; + s);</pre><p>It then connected to the URL, retrieved the response, and processed its contents.</p><p>The important part was:</p><pre>Action action = (Action) Serialize.fromString(string);<br>action.action();<br>serverPrintOut.println(action.output);</pre><p>This revealed the entire mechanism.</p><p>The application:</p><ul><li>Accepted input through the custom service.</li><li>Used that input to construct a URL.</li><li>Retrieved a response from the web server.</li><li>Decoded the Base64 content.</li><li>Deserialized the resulting Java object.</li><li>Executed the object’s action method.</li><li>Returned the command output.</li></ul><p>The Action class was even more interesting.</p><p>It contained a command variable that was passed to a shell process.</p><p>This was a serious security issue.</p><p>The application was deserializing data obtained from a web endpoint and then using the resulting object’s command value in a shell operation.</p><p>In simple terms, the application trusted the serialized object it received from the web server.</p><p>This was the vulnerability that connected everything together.</p><figure><img alt="Analyzing RPG.java revealed how the custom service fetched and deserialized data from the web application." src="https://cdn-images-1.medium.com/max/812/1*nvKmOldQomdqwCp8SS7yWQ.png" /></figure><figure><img alt="" src="https://cdn-images-1.medium.com/max/291/1*CA2bCB7zVCKqfMm86_TeHQ.png" /><figcaption><em>Analyzing RPG.java revealed how the custom service fetched and deserialized data from the web application.</em></figcaption></figure><h3>Understanding the vulnerability chain</h3><p>At this point, the attack path became clear:</p><ol><li>Discover the web application.</li><li>Identify the XXE vulnerability.</li><li>Retrieve the Java source code.</li><li>Understand the unsafe deserialization functionality.</li><li>Follow the intended exploitation path to obtain initial access.</li></ol><p>I now understood why the room had exposed both the web application and the custom service.</p><p>They were part of the same vulnerability chain.</p><h3>6. Getting the Initial Shell</h3><p>After understanding the Java source code, I followed the reference’s approach to construct the expected serialized Java object.</p><p>Luckily, the source code already contained a serialization function:</p><pre>Serialize.toString()</pre><p>This function converted a Java object into a Base64-encoded serialized string.</p><p>The important part was understanding the expected object structure and how the web application reflected its response.</p><p>By connecting the XXE functionality with the custom service’s URL-fetching behavior, I was able to follow the intended exploitation path.</p><p>After completing the interaction, I successfully obtained a reverse shell on the machine.</p><p>Finally, I had my initial foothold!</p><h3>7. Finding the First Clue</h3><p>After obtaining the shell, I started exploring the user’s home directory.</p><p>There was an interesting file named:</p><pre>info.txt</pre><p>I decided to read it.</p><p>The message described a hidden door inside the cave.</p><p>It mentioned that after discovering something unusual about the wall, a wooden door appeared without a keyhole.</p><p>But there was something else.</p><p>The message contained a regular expression, with a hint that the password was hidden inside it.</p><p>This immediately caught my attention.</p><p>A regular expression describes a pattern that strings must follow.</p><p>Since the message specifically mentioned a password, I suspected that the pattern could be used to generate possible password combinations.</p><p>This was another puzzle to solve before progressing.</p><h3>Generating password candidates</h3><p>I searched for ways to generate possible strings from a regular expression and came across a Python utility called <strong>Exrex</strong>.</p><p>Exrex can generate strings that match a supplied regular expression.</p><p>I used the pattern provided in the challenge to generate candidate passwords.</p><p>After generating the candidates, I used Hydra to test the possibilities against the SSH service running on port 2222 in the authorized lab.</p><p>Eventually, I identified the password for the door account.</p><p>Another obstacle cleared!</p><h3>8. Opening the Door</h3><p>With the credentials discovered, I was ready to access the next account.</p><p>Remember that SSH was running on port 2222?</p><p>I used it to connect to the door user:</p><pre>ssh door@10.48.133.224 -p 2222</pre><p>The login was successful!</p><p>I had finally opened the door.</p><h3>9. The Old Man and the Skeleton</h3><p>After logging in as door, I started exploring the home directory.</p><p>I found three interesting files:</p><pre>info.txt<br>oldman.gpg<br>skeleton</pre><p>Naturally, I started by reading info.txt.</p><p>The message described what was waiting ahead:</p><ul><li>An old man speaking through charades.</li><li>A fully armed skeleton blocking the path.</li></ul><p>I decided to execute the skeleton program.</p><p>However, it refused to let me pass.</p><p>The response was:</p><pre>You cannot defeat the skeleton with your current items,<br>your inventory is empty.</pre><p>Interesting!</p><p>The application was checking an inventory before allowing me to continue.</p><p>I remembered something from the earlier Java source-code analysis.</p><p>The serialized lamp object appeared to modify an environment variable.</p><p>I decided to investigate the lamp object again.</p><h3>Understanding the INVENTORY variable</h3><p>After decoding the lamp object, I discovered an interesting command:</p><pre>ls;export INVENTORY=lamp:$INVENTORY</pre><p>This explained how the application maintained the player’s inventory.</p><p>Whenever the lamp action was executed, it added lamp to the INVENTORY environment variable.</p><p>Now I understood why the skeleton was rejecting me.</p><p>The problem wasn’t simply executing the program. I needed to have the correct item in my inventory.</p><p>I also realized that environment variables are associated with a process environment.</p><p>This made me consider whether the way I entered the next account would affect the inventory.</p><h3>10. Finding the Adventurer’s Private Key</h3><p>The next interesting file was:</p><pre>oldman.gpg</pre><p>It was an encrypted file, so I needed to find the corresponding private key and its passphrase.</p><p>I continued enumerating the machine.</p><p>While exploring /var/www, I discovered a directory named:</p><pre>adventurer</pre><p>My current user couldn’t access it, but www-data appeared to have the necessary permissions.</p><p>I remembered something important from RPG.java.</p><p>The application was using the domain:</p><pre>cave.thm</pre><p>This made me wonder whether adventurer was actually configured as a subdomain.</p><p>I investigated the virtual-host configuration and added the discovered hostname to my local hosts file.</p><p>The hostname I wanted to investigate was:</p><pre>adventurer.cave.thm</pre><p>After accessing the virtual host, I discovered a file named:</p><pre>adventurer.priv</pre><p>This appeared to be the private key I needed.</p><p>Finding the private-key passphrase</p><p>Now I had the private key, but there was still one problem.</p><p>When I tried importing it using GPG, it asked for a passphrase.</p><p>I didn’t know the password.</p><p>I went back to the room’s hint and noticed something interesting:</p><p><em>Take a second look in the text files.</em></p><p>I had already read info.txt, but I couldn&#39;t remember seeing anything related to the private-key password.</p><p>So I decided to inspect the file again.</p><p>This time, I used:</p><pre>cat -v info.txt</pre><p>And there it was!</p><p>The file contained a hidden terminal control sequence:</p><pre>^[[A</pre><p>The sequence was affecting how the terminal displayed the text.</p><p>Because of this, part of the information was not immediately visible when reading the file normally.</p><p>Taking a second look revealed the missing private-key passphrase.</p><p>This was a clever little trick in the room.</p><h3>Decrypting the message</h3><p>With the private key and its passphrase, I could finally decrypt the old man’s message.</p><p>I imported the key:</p><pre>gpg --import adventurer.priv</pre><p>Then I decrypted the encrypted file:</p><pre>gpg --output message --no-tty oldman.gpg</pre><p>Finally, I read the message:</p><pre>cat message</pre><p>The message said:</p><pre>IT&#39;S DANGEROUS TO GO ALONE! TAKE THIS [ITEM]</pre><p>A familiar reference!</p><p>The old man had finally given me the item needed to defeat the skeleton.</p><h3>11. Defeating the Skeleton</h3><p>Now that I had the required item, I needed to add it to my inventory.</p><p>There was one important detail I remembered from the lamp object.</p><p>The inventory was maintained through an environment variable.</p><p>I set the inventory accordingly and executed the skeleton program again.</p><p>This time, it returned credentials for the next account.</p><p>The skeleton was finally defeated!</p><h3>Switching to the skeleton account</h3><p>I used the credentials returned by the program to switch users:</p><pre>su skeleton</pre><p>This was an important distinction from connecting through SSH.</p><p>The inventory was maintained through environment variables, so switching users and starting a fresh SSH session could produce different results.</p><p>After switching successfully, I explored the new home directory.</p><p>There was another info.txt file.</p><p>I decided to read it.</p><p>The message described a huge magical tree.</p><p>I could see the tree, but an invisible wall prevented me from reaching its root.</p><p>The cave was telling me that I was close to freedom, but there was still another obstacle.</p><h3>12. Discovering the Privilege Escalation Path</h3><p>Now it was time for more Linux enumeration.</p><p>I started by checking my sudo permissions:</p><pre>sudo -l</pre><p>The output revealed something interesting.</p><p>The skeleton user could execute the following command as root without a password:</p><pre>/bin/kill</pre><p>At first, I wasn’t sure how useful this would be.</p><p>I checked the usual privilege-escalation references, but nothing immediately suggested an obvious route.</p><p>So I continued exploring.</p><h3>The unusual directory under /opt</h3><p>While inspecting the filesystem, I discovered a directory named:</p><pre>/opt/link</pre><p>I listed its contents:</p><pre>ls -la</pre><p>There was a symbolic link:</p><pre>startcon -&gt; ../root/start.sh</pre><p>This immediately caught my attention.</p><p>The link pointed toward a startup script inside the root directory.</p><p>I decided to investigate the link itself.</p><p>After moving it to /tmp, I inspected its contents.</p><p>The script contained:</p><pre>#!/bin/bash</pre><pre>service ssh start<br>service apache2 start<br>su - cave -c &quot;cd /home/cave/src; ./run.sh&quot;</pre><pre>/bin/bash</pre><p>Now things were getting interesting.</p><p>The script was responsible for starting SSH, Apache, and the cave application.</p><p>More importantly, I discovered that the script was writable.</p><h3>13. Discovering Docker</h3><p>Before going any further, I wanted to understand the environment I was operating in.</p><p>I checked the following file:</p><pre>cat /proc/1/cgroup</pre><p>The output revealed that I was inside a Docker container.</p><p>This explained why the startup script and process-management behavior were so important.</p><p>The room was no longer just a Linux privilege-escalation challenge.</p><p>It was also testing my understanding of container isolation.</p><h3>Understanding the container restart behavior</h3><p>Since the startup script was writable, I investigated how the container initialized its services and what happened when its main process stopped.</p><p>I learned that the process with PID 1 plays a special role inside a container.</p><p>It is responsible for the container’s main process lifecycle.</p><p>The unusual sudo permission involving /bin/kill now made much more sense.</p><p>The challenge was giving me a way to interact with process management, while the writable startup script provided another important piece of the puzzle.</p><p>I followed the intended room methodology and used the startup behavior to trigger a connection back to my lab machine.</p><p>Eventually, I received a reverse shell from the next environment.</p><p>This was another major milestone.</p><p>I had moved beyond the original cave application environment.</p><h3>14. Escaping the Docker Container</h3><p>After receiving the next shell, I explored the environment again.</p><p>The room’s storyline described reaching the root of a magical tree.</p><p>However, the message also hinted that I was still contained inside the cave.</p><p>That description was actually quite accurate.</p><p>I was inside a Docker container, and I needed to understand how to reach the underlying host.</p><h3>Understanding privileged containers</h3><p>Docker containers use Linux isolation mechanisms to separate their processes and resources from the host.</p><p>However, a container configured with excessive privileges can weaken these security boundaries.</p><p>During enumeration, I discovered that the container had access to privileged functionality involving Linux control groups.</p><p>This brought me to the concept of the <strong>cgroup release-agent mechanism</strong>.</p><h3>What is the cgroup release-agent mechanism?</h3><p>Linux control groups, or cgroups, are used to organize processes and manage system resources.</p><p>The release-agent mechanism can execute a configured program when a cgroup becomes empty under the appropriate conditions.</p><p>In an incorrectly configured privileged container, access to this mechanism can create a serious security issue.</p><p>If the container can influence the host-side release-agent configuration and the host executes the resulting program, the isolation boundary can be broken.</p><p>This was the vulnerability I needed to understand for the final stage.</p><p>I followed the intended challenge path and successfully obtained access to the underlying host.</p><p>The cave was finally opening up!</p><h3>Reaching the final root directory</h3><p>After obtaining host-level access, I navigated to:</p><p>bash</p><pre>cd /root</pre><p>Inside the directory, I discovered another info.txt file.</p><p>I decided to read it.</p><p>The final message described digging beneath the magical tree, discovering that its roots continued downward, and eventually reaching the outside world.</p><p>It was a fitting conclusion to the entire room.</p><p>I had finally escaped the cave.</p><p>The final flag was waiting for me.</p><h3>15. What I Learned</h3><p>This room was a great example of how multiple vulnerabilities and misconfigurations can be chained together.</p><p>There were several important lessons I took away from this challenge.</p><h3>1. XXE can reveal more than just files</h3><p>The XXE vulnerability was the starting point for understanding the application’s internal logic.</p><p>It helped expose the Java source code, which revealed how the custom service processed user-controlled data.</p><h3>2. Always investigate unusual application behavior</h3><p>The service on port 3333 initially looked like a simple text-based game.</p><p>But understanding how it communicated with the web server revealed a much more serious security issue.</p><h3>3. Unsafe Java deserialization is dangerous</h3><p>Deserializing untrusted data can introduce serious security risks.</p><p>Developers should avoid unsafe deserialization and never allow untrusted objects to influence sensitive operations.</p><h3>4. Environment variables can be important</h3><p>The INVENTORY variable was a clever part of the challenge.</p><p>Understanding how environment variables are inherited and modified helped me understand why switching users with su mattered.</p><h3>5. Hidden terminal characters can reveal information</h3><p>The private-key passphrase was hidden behind a terminal control sequence.</p><p>Using cat -v was a simple but effective way to investigate the file more carefully.</p><h3>6. Docker isolation depends on configuration</h3><p>The final stages demonstrated why privileged containers and excessive capabilities are dangerous.</p><p>A container should not automatically be trusted simply because it provides process isolation.</p><h3>Conclusion</h3><p>Honestly, this was one of the more interesting TryHackMe rooms I’ve completed.</p><p>What started with a simple question on port 3333 turned into a journey involving web exploitation, Java deserialization, Linux privilege escalation, encryption, and Docker security.</p><p>My favorite part was how the different clues connected to one another.</p><p>The matches helped me understand the environment, the lamp revealed the Java application files, the old man provided the missing item, and the skeleton introduced another layer of Linux privilege escalation.</p><p>The final Docker escape was a great way to bring everything together.</p><p>More importantly, this room reminded me that penetration testing isn’t just about running tools and finding vulnerabilities.</p><p>It’s about understanding how applications work, investigating unusual behavior, connecting discoveries, and thinking logically about the environment.</p><p>If you’re interested in web exploitation, Linux privilege escalation, and container security, I definitely recommend giving this room a try.</p><p><strong>Thanks for reading, and happy hacking! 🚀</strong></p><h3>Do u get this answers ???</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/567/1*TRthpBfK-la3u1jdNgNLPg.png" /></figure><h3>Credits and References</h3><ul><li>TryHackMe — You’re in a Cave</li><li>Nmap documentation</li><li>Gobuster documentation</li><li>Exrex documentation</li><li>Hydra documentation</li><li>GnuPG documentation</li><li>Docker security documentation</li><li>Linux cgroups documentation</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b43fb30dd553" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[AI-Powered Cyberattacks Are Targeting India’s Financial Sector: A Threat Intelligence Analyst’s…]]></title>
            <link>https://medium.com/@trivediaarsh51/ai-powered-cyberattacks-are-targeting-indias-financial-sector-a-threat-intelligence-analyst-s-1a6ea21e0726?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/1a6ea21e0726</guid>
            <category><![CDATA[payment-api-abuse]]></category>
            <category><![CDATA[soc-detection]]></category>
            <category><![CDATA[mitre-attack-framework]]></category>
            <category><![CDATA[ai-offensive-security]]></category>
            <category><![CDATA[wireshark]]></category>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Sat, 07 Nov 2026 06:51:09 GMT</pubDate>
            <atom:updated>2026-11-07T06:51:09.805Z</atom:updated>
            <content:encoded><![CDATA[<h3>AI-Powered Cyberattacks Are Targeting India’s Financial Sector: A Threat Intelligence Analyst’s Deep Dive into CERT-In’s Crafty Falcon Warning</h3><p><em>How AI-assisted offensive tools and exposed APIs in security mechanisms are changing the financial-sector’s threat landscape.</em></p><h3>1. Introduction: The Changing Nature of Cyber Threats</h3><p>Artificial Intelligence is transforming the cybersecurity landscape.</p><p>While organizations are increasingly adopting AI to improve threat detection, automate security operations and strengthen incident response, threat actors are also finding ways to use AI-assisted tools to accelerate their operations.</p><p>On November 5, 2026, Reuters reported on a significant warning from India’s Computer Emergency Response Team (CERT-In) concerning an AI-assisted cyberattack campaign targeting financial institutions.</p><p>CERT-In’s November 4 advisory, identified as <strong>CIAD-2026–0058</strong>, designated the threat actor as <strong>CRAFTY FALCON</strong>.</p><p>The campaign reportedly targets:</p><ul><li>Non-Banking Financial Companies (NBFCs).</li><li>Digital lending platforms.</li><li>Fintech organizations.</li><li>Payment aggregators and payment service providers.</li><li>Cryptocurrency exchanges and virtual digital asset service providers.</li></ul><p>According to CERT-In, the campaign remains active and affected organizations have experienced repeated targeting.</p><p><strong>Primary source:</strong> <a href="https://www.cert-in.org.in/s2cMainServlet?VLCODE=CIAD-2026-0058&amp;pageid=PUBVLNOTES02">CERT-In Advisory CIAD-2026–0058</a></p><h3>2. Understanding the CRAFTY FALCON Campaign</h3><p>CERT-In describes CRAFTY FALCON as an AI-assisted financial threat actor that surveys targets using AI and invokes victim’s own payment and API’s.</p><p>The advisory identifies four tools associated with the campaign:</p><p>Tool Description:</p><ul><li>ARTEX : LLM-based offensive security tool family</li><li>PentAGI : Autonomous multi-agent AI penetration testing system</li><li>Cairn : Attack search engine</li><li>CyberStrikeAI : AI-assisted offensive security framework</li></ul><p>This introduces an important challenge for security teams.</p><p>Traditional security monitoring often depends on identifying recognizable patterns of suspicious activity. However, automated operations can generate numerous related events in a short period, making manual investigation more difficult.</p><p>For threat intelligence analysts, this raises several important questions:</p><ul><li>What infrastructure and services are being targeted?</li><li>What weaknesses are being exploited?</li><li>Are attackers abusing legitimate application functionality?</li><li>What indicators can help defenders identify suspicious activity?</li><li>How can organizations identify repeated attacks after remediation?</li></ul><p>These questions are central to understanding and defending against modern financial-sector cyber threats.</p><figure><img alt="Crafty Falcon — CERT-IN" src="https://cdn-images-1.medium.com/max/1000/0*IWrop-Vj_NOWdwsW" /><figcaption>Crafty Falcon — CERT-IN</figcaption></figure><h3>3. Understanding the Reported Attack Pattern</h3><p>CERT-In’s November advisory describes several important characteristics of the campaign.</p><h3>3.1 Exploitation of exposed environments</h3><p>Internet-facing applications and non-production environments, including staging, testing and UAT systems, can provide opportunities for unauthorized access.</p><p>A particularly concerning scenario involves attackers gaining access to non-production environments and subsequently reaching production systems through shared credentials, tokens or network paths.</p><p>From a Blue Team perspective, this highlights the importance of separating development, testing and production environments.</p><h3>3.2 Abuse of legitimate payment APIs</h3><p>One of the most important observations in the advisory concerns the abuse of payment and APIs.</p><p>Financial organizations use APIs to connect applications, payment providers, banks and other services.</p><p>If attackers obtain access to API credentials or exploit weaknesses in application; they may attempt to invoke legitimate payment functionality without following the organization’s intended business workflow.</p><p>This demonstrates why API security cannot depend exclusively on trusting requests from an authorized application.</p><p>Server-side authorization and transaction validation must independently verify every sensitive operation.</p><h3>3.3 Repeated attacks after incomplete remediation</h3><p>Another concerning aspect is the reported retargeting of organizations that had previously undertaken remediation.</p><p>This highlights an important lesson:</p><p><strong>Fixing the initial vulnerability does not necessarily mean the entire attack path has been eliminated.</strong></p><p>Security teams must investigate:</p><ul><li>Whether compromised credentials remain valid.</li><li>Whether API tokens have been rotated.</li><li>Whether unauthorized persistence remains.</li><li>Whether staging and production environments are properly isolated.</li><li>Whether all affected applications and integrations have been reviewed.</li></ul><h3>3.4 Attacks during low-monitoring periods</h3><p>CERT-In also reports that attackers timed operations around weekends, holidays and other periods with reduced monitoring coverage.</p><p>This creates an operational challenge for SOC teams.</p><figure><img alt="Conceptual Representation of Analyst" src="https://cdn-images-1.medium.com/max/1000/0*Avr1ca4m-rujI-mI" /><figcaption>Conceptual Representation of Analyst</figcaption></figure><h3>4. Threat Intelligence Analysis: What Should Analysts Investigate?</h3><p>Threat intelligence is not simply collecting malicious IP addresses or publishing lists of indicators of compromise.</p><p>It involves understanding adversary behavior, identifying relevant evidence, assessing threats and communicating intelligence that helps organizations make better security decisions.</p><p>For this campaign, I would divide the investigation into four areas.</p><h3>4.1 Infrastructure intelligence</h3><p>Analysts should investigate:</p><ul><li>Suspicious source IP addresses.</li><li>Hosting-provider infrastructure.</li><li>Domain registration patterns.</li><li>DNS resolution history.</li><li>TLS certificate characteristics.</li><li>Infrastructure reuse across separate incidents.</li></ul><p>CERT-In reports that supporting attack infrastructure can rotate IP addresses within hours to days.</p><p>This makes relying exclusively on static IP blocklists insufficient.</p><p>Behavioral indicators and infrastructure relationships should also be considered.</p><h3>4.2 Tooling intelligence</h3><p>The publicly identified tools provide another area for investigation.</p><p>Security teams should monitor for:</p><ul><li>Unauthorized offensive security tool downloads.</li><li>Unexpected AI coding-agent activity.</li><li>Unusual LLM API usage.</li><li>Suspicious processes executing on corporate servers.</li><li>Unexpected outbound connections from non-development systems.</li></ul><p>An important distinction is that legitimate security teams may also use penetration-testing and AI-assisted security tools.</p><p>Therefore, tool detection alone should not automatically classify an event as malicious.</p><p>Authorization, execution context, user identity and system behavior must be considered.</p><h3>4.3 Identity and credential intelligence</h3><p>Credential security is particularly important in financial environments.</p><p>Analysts should investigate:</p><ul><li>Unusual API authentication patterns.</li><li>Unexpected use of service accounts.</li><li>Privileged access outside approved workflows.</li><li>Reuse of credentials across environments.</li><li>Access involving expired or previously rotated credentials.</li></ul><p>A valid authentication event does not necessarily represent legitimate activity.</p><h3>4.4 Transaction intelligence</h3><p>Financial organizations should also correlate security events with business transactions.</p><p>For example:</p><ul><li>Unusual payment frequencies.</li><li>Unexpected beneficiary changes.</li><li>Transactions outside normal operating patterns.</li><li>Repeated low-value transactions followed by unusual activity.</li><li>API requests that bypass expected approval workflows.</li></ul><p>This is where collaboration between cybersecurity teams, fraud teams and application owners becomes particularly valuable.</p><h3>5. MITRE ATT&amp;CK Mapping: A Defensive Perspective</h3><p>MITRE ATT&amp;CK provides a common language for describing adversary behavior.</p><p>The following table represents potential technique mappings for investigating the reported attack pattern.</p><p>These are investigation hypotheses, not confirmation that every technique was used by CRAFTY FALCON.</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fdocs.google.com%2Fdocument%2Fd%2F1cJR5A0YHJHOzUwKzWQJZkGRmAFrcXfzFSoZ97Z1Z2G0%2Fpreview%3Fembedded%3Dtrue&amp;display_name=Google+Docs&amp;url=https%3A%2F%2Fdocs.google.com%2Fdocument%2Fd%2F1cJR5A0YHJHOzUwKzWQJZkGRmAFrcXfzFSoZ97Z1Z2G0%2Fedit%3Fusp%3Dsharing&amp;image=https%3A%2F%2Flh7-us.googleusercontent.com%2Fdocs%2FAHkbwyKkI17s9vO82JXrvPQyys0DzD7l6I2YzTaQtJF47p0bfI-5uabcSOYHtuyuZDZmt5d8_o7KNR1VMZDZAg2dCTafptCUcgOzW6gaAjZ-tSOoIGph9rsX%3Dw1200-h630-p&amp;type=text%2Fhtml&amp;scroll=auto&amp;schema=google" width="700" height="530" frameborder="0" scrolling="no"><a href="https://medium.com/media/e1ea7ea0921081a6ff5fc4ed8deb5e3b/href">https://medium.com/media/e1ea7ea0921081a6ff5fc4ed8deb5e3b/href</a></iframe><p>Reference: MITRE ATT&amp;CK Enterprise Matrix</p><h3>Why this mapping matters</h3><p>MITRE ATT&amp;CK helps analysts translate technical observations into a structured understanding of adversary behavior.</p><p>For example, if an organization discovers unauthorized access through an internet-facing application, the analyst can investigate whether the activity aligns with T1190.</p><p>If the investigation subsequently identifies suspicious use of compromised credentials, T1078 may be relevant.</p><p>However, analysts should never assign techniques merely because they seem plausible.</p><p>Every mapping should be supported by telemetry, forensic evidence or reliable reporting.</p><h3>6. SOC Detection Strategy: What Should Security Teams Monitor?</h3><p>A modern SOC should correlate multiple sources of telemetry rather than depend on a single alert.</p><p>For this campaign, I would prioritize five detection areas.</p><h3>Detection Area 1: Suspicious API activity</h3><p>Monitor for:</p><ul><li>Unexpected API request volumes.</li><li>Unusual authentication patterns.</li><li>API requests from unfamiliar infrastructure.</li><li>Unusual service-account activity.</li><li>Transaction requests that bypass expected approval processes.</li></ul><h3>Detection Area 2: Authentication anomalies</h3><p>Investigate:</p><ul><li>Repeated failed authentication attempts.</li><li>Successful logins following unusual failure patterns.</li><li>Unexpected service-account access.</li><li>Access during unusual operating periods.</li></ul><h3>Detection Area 3: Cloud management activity</h3><p>CERT-In specifically identifies potential abuse of cloud management services, including AWS Systems Manager and Lambda.</p><p>Security teams should review:</p><ul><li>Unexpected administrative actions.</li><li>Unusual command execution through management services.</li><li>Unexpected deployment of files.</li><li>Changes to cloud permissions.</li><li>Activity originating from unfamiliar identities.</li></ul><h3>Detection Area 4: Unauthorized AI tooling</h3><p>Monitor:</p><ul><li>Unexpected LLM API usage.</li><li>Unapproved AI-assisted security tools.</li><li>Unusual processes on production servers.</li><li>Unexpected outbound communication from non-development systems.</li></ul><h3>Detection Area 5: Repeated suspicious activity</h3><p>A particularly useful detection strategy is identifying activity against systems that were previously involved in a security incident.</p><p>Correlating new alerts with historical incident records may reveal incomplete remediation or recurring intrusion attempts.</p><figure><img alt="SOC investigation workflow" src="https://cdn-images-1.medium.com/max/800/0*OHS_F70T_sIlkjDk" /><figcaption>SOC investigation workflow</figcaption></figure><h3>7. Practical Investigation: A Beginner-Friendly Blue Team Lab</h3><p>To understand how analysts can investigate suspicious network activity, I would build a small defensive lab using Wireshark and a simulated dataset.</p><p><strong>Important:</strong> This is a proposed educational simulation. It is not a reproduction of the CRAFTY FALCON campaign, and the following events are not actual CERT-In indicators.</p><h3>Lab objective</h3><p>Investigate suspicious network and authentication activity and produce a short threat intelligence report.</p><h3>Tools</h3><ul><li>Wireshark — network traffic analysis.</li><li>Kali Linux — isolated security lab environment.</li></ul><h3>Step 1: Prepare a safe lab</h3><p>Use an isolated virtual environment and a packet capture from your own lab or an authorized training exercise.</p><p>Do not scan or test financial institutions, public APIs or third-party infrastructure.</p><h3>Step 2: Investigate network traffic using Wireshark</h3><p>Start by examining DNS and TLS metadata.</p><p>Useful Wireshark display filters:</p><pre>dns</pre><p>To inspect TLS server names:</p><pre>tls.handshake.extensions_server_name</pre><p>To inspect TCP connection initiation:</p><pre>tcp.flags.syn == 1 &amp;&amp; tcp.flags.ack == 0</pre><p>To investigate a particular connection:</p><pre>tcp.stream eq 0</pre><p>These filters help analysts examine DNS requests, visible TLS handshake metadata and TCP communication patterns.</p><p>One limitation is important: HTTPS application content is encrypted unless authorized decryption is available. Wireshark cannot reveal the contents of encrypted payment API requests simply by filtering on port 443.</p><h3>Step 3: Examine simulated security events</h3><p>Consider this fictional example:</p><iframe src="https://cdn.embedly.com/widgets/media.html?src=https%3A%2F%2Fdocs.google.com%2Fspreadsheets%2Fd%2F1RlnzpMpl4dDAAfHeVXoEDygOvZ2mX6CeAR5RxzbEXcw%2Fpreview%3Fembedded%3Dtrue&amp;display_name=Google+Docs&amp;url=https%3A%2F%2Fdocs.google.com%2Fspreadsheets%2Fd%2F1RlnzpMpl4dDAAfHeVXoEDygOvZ2mX6CeAR5RxzbEXcw%2Fedit%3Fusp%3Dsharing&amp;image=https%3A%2F%2Flh7-us.googleusercontent.com%2Fdocs%2FAHkbwyKBe3DNEGohcsR6iB1fD7KP9E_g12ZTfNwQfq3Wbbq0VyQZV-AxRCi7I2cy7uDvEgNTEEJFYXBK6vvpaPAm5yBTiNJaABsuJH6M8-yOQjIu-G4D_Efv%3Dw1200-h630-p&amp;type=text%2Fhtml&amp;scroll=auto&amp;schema=google" width="700" height="530" frameborder="0" scrolling="no"><a href="https://medium.com/media/f39538d89cbfe191ce26d48952337b08/href">https://medium.com/media/f39538d89cbfe191ce26d48952337b08/href</a></iframe><p>The analyst should not immediately declare this a confirmed attack.</p><p>Instead, investigate:</p><ol><li>Was the account authorized to access the API?</li><li>Was the authentication source expected?</li><li>Did the activity match normal business behavior?</li><li>Were unusual transactions approved?</li><li>Were related security alerts generated?</li><li>Did the activity originate from a previously compromised system?</li></ol><h3>Step 4: Develop a detection hypothesis</h3><p>A useful initial detection hypothesis could be:</p><p>“An unusual increase in payment API requests, combined with suspicious authentication activity and unexpected transaction behavior, may indicate unauthorized use of financial application functionality.”</p><p>This hypothesis can be translated into SIEM correlation logic.</p><p>For example, a conceptual detection rule:</p><pre>IF<br>    authentication_activity = suspicious<br>AND<br>    payment_api_activity = unusually_high<br>AND<br>    transaction_behavior = anomalous<br>THEN<br>    generate_high_priority_alert<br>    investigate_account_and_source</pre><p>This is illustrative pseudocode, not a production-ready SIEM rule.</p><p>A production implementation should use baselines, thresholds, allowlists, identity context and transaction validation to reduce false positives.</p><h3>Step 5: Write an investigation report</h3><p>A professional SOC investigation should contain:</p><ul><li>Alert description.</li><li>Affected system.</li><li>Relevant timestamps.</li><li>User or service-account context.</li><li>Network observations.</li><li>Supporting log evidence.</li><li>MITRE ATT&amp;CK hypotheses.</li><li>Investigation findings.</li><li>Recommended containment actions.</li></ul><p>This is the type of documentation that demonstrates practical analytical thinking.</p><h3>8. Recommendations for Financial Organizations</h3><p>Based on the reported attack patterns, I would highlight the following defensive recommendations.</p><p><strong>1. Strengthen API security</strong></p><p>Implement server-side authorization, strict transaction validation, rate limiting and independent approval mechanisms.</p><p><strong>2. Improve credential management</strong></p><p>Rotate potentially compromised credentials, protect API secrets and avoid sharing credentials between environments.</p><p><strong>3. Separate environments</strong></p><p>Maintain clear security boundaries between development, testing, staging and production.</p><p><strong>4. Improve cloud monitoring</strong></p><p>Monitor administrative operations, unusual service executions, permission changes and unexpected file deployments.</p><p><strong>5. Strengthen SOC coverage</strong></p><p>Ensure critical security alerts receive appropriate attention during weekends, holidays and low-monitoring periods.</p><p><strong>6. Improve incident remediation</strong></p><p>Validate the complete attack path instead of closing incidents after addressing only the initial vulnerability.</p><p><strong>7. Correlate cyber and financial events</strong></p><p>Connect application security telemetry with payment and transaction monitoring to identify suspicious activity earlier.</p><h3>9. What This Means for the Future of Threat Intelligence</h3><p>The CRAFTY FALCON warning highlights an important development in cybersecurity: AI-assisted tooling can increase the speed and scale of offensive activity.</p><p>However, the defensive challenge is not simply detecting AI.</p><p>It is understanding the behavior of an attacker, identifying weaknesses in organizational controls and recognizing suspicious activity across multiple systems.</p><p>For threat intelligence analysts, this means developing skills in:</p><ul><li>Open-source intelligence (OSINT).</li><li>Threat actor behavior analysis.</li><li>MITRE ATT&amp;CK.</li><li>Indicator of Compromise (IOC) enrichment.</li><li>Network traffic analysis.</li><li>SIEM investigation.</li><li>Cloud security monitoring.</li><li>Intelligence reporting.</li></ul><p>It also reinforces the importance of distinguishing confirmed evidence from assumptions.</p><p>Attribution should be based on reliable intelligence rather than the perceived origin of a tool or its developer.</p><figure><img alt="Threat intelligence lifecycle" src="https://cdn-images-1.medium.com/max/1000/0*yAv62cT9y7GFu48U" /><figcaption>Threat intelligence lifecycle</figcaption></figure><h3>10. My Takeaway as an Aspiring Threat Intelligence Analyst</h3><p>As someone beginning a journey in Blue Team cybersecurity, I find this development particularly interesting because it demonstrates how quickly the threat landscape is evolving.</p><p>The most important lesson I take from this advisory is that organizations cannot depend solely on traditional perimeter security or static indicators of compromise.</p><p>They need continuous monitoring, strong identity controls, secure APIs, effective incident response and threat intelligence that connects technical activity with business risk.</p><p>I also believe aspiring cybersecurity professionals should move beyond simply following security news.</p><p>We should learn to investigate it, understand the attack patterns, identify detection opportunities and communicate practical defensive recommendations.</p><p>That is the kind of analytical thinking I want to continue developing as I work toward a career in threat intelligence.</p><p><strong>The question I leave readers with is:</strong></p><p><em>If AI-assisted attacks can operate at machine speed, how should organizations evolve their threat intelligence and SOC capabilities to detect and respond before financial damage occurs?</em></p><p>I would love to hear your thoughts.</p><h3>References and Further Reading</h3><ol><li><strong>CERT-In — Advisory CIAD-2026–0058 (November 4, 2026):</strong> <a href="https://www.cert-in.org.in/s2cMainServlet?VLCODE=CIAD-2026-0058&amp;pageid=PUBVLNOTES02">Escalation of AI-Driven Attacks on BFSI Entities by Threat Actor CRAFTY FALCON</a></li><li><strong>Reuters — November 5, 2026:</strong> <a href="https://www.reuters.com/legal/government/india-flags-chinese-linked-ai-tools-hacking-campaign-against-financial-firms-2026-11-05/">India flags AI tools in hacking campaign against financial firms</a></li><li><strong>CERT-In — October 9, 2026:</strong> Threat Actors Exploiting Payment and Disbursement APIs for Unauthorized Fund Transfers</li><li><strong>MITRE ATT&amp;CK:</strong> Enterprise Adversary Tactics and Techniques</li><li><strong>Wireshark:</strong> Official Documentation</li></ol><p><em>Disclaimer: This article is an independent educational analysis based on publicly available reporting and CERT-In advisories. The practical investigation is a hypothetical defensive exercise. It does not claim access to CERT-In’s restricted indicators, victim telemetry or forensic findings.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=1a6ea21e0726" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Inside the European Commission Data Breach: How a Supply-Chain Attack Turned Into a Cloud Security…]]></title>
            <link>https://medium.com/@trivediaarsh51/inside-the-european-commission-data-breach-how-a-supply-chain-attack-turned-into-a-cloud-security-49b388b496a1?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/49b388b496a1</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Wed, 04 Nov 2026 16:37:10 GMT</pubDate>
            <atom:updated>2026-11-04T16:37:10.778Z</atom:updated>
            <content:encoded><![CDATA[<h3>Inside the European Commission Data Breach: How a Supply-Chain Attack Turned Into a Cloud Security Incident</h3><p><em>An incident investigation into compromised software, stolen AWS credentials, data exfiltration, and the lessons every security team should take seriously.</em></p><p>On March 24, 2026, the European Commission’s Cybersecurity Operations Centre detected suspicious AWS API activity, possible account compromise, and an unexpected increase in network traffic.</p><p>An attacker had reportedly obtained cloud credentials through a compromised software supply chain; used those credentials to access European Commission infrastructure, and exfiltrated a significant volume of data.</p><p>By March 28, the stolen data had been published on a dark web leak site.</p><p>This incident is a reminder that even organizations with dedicated security teams can be affected by vulnerabilities far beyond their immediate infrastructure.</p><h3>1. Incident overview</h3><p>The incident affected the European Commission’s Europa web hosting platform, which runs on Amazon Web Services (AWS).</p><p>According to the official investigation by CERT-EU:</p><ul><li>Approximately 91.7 GB of compressed data was exfiltrated.</li><li>The exposed information included names, usernames, email addresses, and potentially email content.</li><li>The affected hosting service supported up to 71 clients, including 42 European Commission clients and at least 29 other EU entities.</li><li>The attacker obtained an AWS secret through a software supply-chain compromise involving Trivy.</li><li>The stolen dataset was publicly released by the ShinyHunters extortion group.</li></ul><h3>2. How did the attack happen?</h3><p>The initial attack was linked with high confidence to the Trivy supply-chain compromise attributed to TeamPCP.</p><p>Trivy is a widely used security scanner that helps organizations identify vulnerabilities and misconfigurations in software and infrastructure.</p><p>The danger in a supply-chain attack is that organizations may trust software because it comes through a legitimate distribution channel. If that software is compromised, the trust placed in it can become an attacker’s opportunity.</p><h3>Step 1: Compromised software reaches the environment</h3><p>The European Commission was using a compromised version of Trivy during the relevant period.</p><p>According to the investigation, the malicious software enabled the attacker to obtain an AWS API key on March 19, 2026.</p><p>That key had privileges over other AWS accounts associated with the Commission.</p><h3>Step 2: Credential discovery and abuse</h3><p>After obtaining the AWS secret, the attacker attempted to discover additional credentials using TruffleHog, a tool commonly used for secret scanning.</p><p>The attacker also created a new access key for an existing user, apparently to maintain access while making the activity less conspicuous.</p><p>This is a critical point: the attacker did not need to break through every cloud security control. They were able to operate using credentials that AWS recognized as valid.</p><h3>Step 3: Cloud reconnaissance</h3><p>With the compromised credentials, the attacker began exploring the cloud environment.</p><p>The investigation confirmed that the compromised account had management rights that could potentially have enabled access to other AWS accounts.</p><p>Although this created a risk of lateral movement, investigators had not found evidence that such movement occurred.</p><h3>Step 4: Data exfiltration</h3><p>The attacker accessed and extracted data from the affected AWS environment.</p><p>The compromised infrastructure supported multiple websites hosted through the Europa service. As a result, the potential exposure extended beyond one website or one organization.</p><p>The dataset was approximately 91.7 GB compressed, or around 340 GB uncompressed.</p><h3>Step 5: Public leak</h3><p>On March 28, 2026, ShinyHunters published the stolen dataset on its dark web leak site.</p><p>The publication transformed a cloud compromise into a wider data exposure incident, increasing the potential for phishing, impersonation, and misuse of personal information.</p><h3>3. What data was exposed?</h3><p>The investigation confirmed that the leaked dataset contained personal information, including:</p><ul><li>Names and surnames</li><li>Usernames</li><li>Email addresses</li><li>Email-related files, some of which could contain original user-submitted messages</li></ul><p>The dataset included at least 51,992 files associated with outbound email communications, totaling approximately 2.22 GB.</p><p>Most were automated notifications with little content. However, some bounce-back messages could contain the original emails submitted by users.</p><p>The analysis of databases associated with the hosted websites was still ongoing when CERT-EU published its report.</p><p>This distinction matters. A forensic investigation should clearly separate confirmed exposure from information that remains under examination.</p><h3>4. Incident response and containment</h3><p>The Commission and CERT-EU took several steps to contain the incident and reduce further risk.</p><h3>Credential containment</h3><p>The compromised AWS secret was secured, and access keys created or abused by the attacker were disabled or deleted.</p><p>This was essential because stolen credentials can remain useful even after the original software vulnerability has been addressed.</p><h3>Investigation and impact assessment</h3><p>Investigators examined the compromised cloud environment, analyzed the exposed dataset, and assessed whether other cloud accounts or hosted organizations might have been affected.</p><h3>Regulatory and stakeholder notification</h3><p>The Commission notified relevant data protection authorities and began communicating directly with affected hosting clients.</p><h3>Continued monitoring</h3><p>The Commission confirmed that it would continue monitoring its systems and sharing additional findings as the investigation progressed.</p><p>The response illustrates why incident handling involves more than removing malicious software. It also requires credential revocation, forensic analysis, impact assessment, regulatory coordination, and communication.</p><h3>6. Root causes and security weaknesses</h3><p>The incident was not caused by a single isolated failure. It exposed several interconnected risks.</p><h3>Software supply-chain trust</h3><p>A compromised security tool became the route through which cloud credentials were exposed.</p><p>Organizations must recognize that security tools themselves can become attack vectors.</p><h3>Excessive credential privileges</h3><p>The compromised AWS secret had management rights over other cloud accounts.</p><p>This increased the potential impact of the compromise. Least-privilege access could have reduced the attacker’s ability to operate across the environment.</p><h3>Cloud credential exposure</h3><p>Credentials accessible to software and CI/CD environments can become valuable targets.</p><p>Secrets should be tightly scoped, monitored, and rotated whenever exposure is suspected.</p><h3>Detection and investigation challenges</h3><p>The attacker used valid credentials and created additional access keys. Such activity can resemble legitimate administrative operations.</p><p>This makes behavioral monitoring and detailed audit logs essential.</p><h3>References and further reading</h3><ol><li><a href="https://cert.europa.eu/blog/european-commission-cloud-breach-trivy-supply-chain">CERT-EU: European Commission cloud breach — a supply-chain compromise</a></li><li><a href="https://ec.europa.eu/commission/presscorner/detail/en/ip_26_748">European Commission: Official incident announcement</a></li><li><a href="https://www.aquasec.com/">Aqua Security: Trivy supply-chain attack advisory</a></li><li><a href="https://attack.mitre.org/techniques/T1195/002/">MITRE ATT&amp;CK: Supply Chain Compromise (T1195.002)</a></li><li><a href="https://attack.mitre.org/techniques/T1078/004/">MITRE ATT&amp;CK: Valid Accounts — Cloud Accounts (T1078.004)</a></li></ol><p><em>This article is an independent OSINT analysis based on publicly available incident reporting.</em></p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=49b388b496a1" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[OSINT CTF Walkthrough]]></title>
            <link>https://medium.com/@trivediaarsh51/osint-ctf-walkthrough-4e7c46f61c76?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e7c46f61c76</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Thu, 27 Aug 2026 12:06:57 GMT</pubDate>
            <atom:updated>2026-08-27T12:06:57.908Z</atom:updated>
            <content:encoded><![CDATA[<h3>From a Hidden .onion Link to an iPhone 7 Plus : Intro-To-Chinese-Osint by OSINTUK</h3><blockquote><strong><em>Disclaimer:</em></strong><em> This article is a walkthrough of a fictional/CTF investigation. The techniques discussed are intended for authorized OSINT research, CTFs, digital investigations, and security research. Do not use these techniques to obtain or expose private information about real individuals without authorization.</em></blockquote><p>OSINT investigations rarely follow a straight line.</p><p>You start with one piece of information — a username, domain, phone number, image, or social-media account — and gradually pivot between completely different platforms until seemingly unrelated pieces of information begin to connect.</p><p>This CTF was a great example of exactly that.</p><p>The investigation started with a <strong>Chinese blog post containing a hidden </strong><strong>.onion address</strong> and eventually moved through:</p><p><strong>Chinese blog → Tor domain → Telegram → QQ → phone number → company domain → Wayback Machine → secondary QQ → phone number → Weibo → historical device metadata</strong></p><p>In total, the challenge consisted of <strong>13 interconnected OSINT challenges</strong>.</p><p>This is a complete walkthrough of the investigation and, more importantly, the OSINT techniques used at each stage.</p><h3>Challenge 1 — Finding the Hidden .onion Link</h3><h3>Objective</h3><p>Find a hidden Tor .onion link embedded inside a Chinese blog post.</p><h3>The Problem</h3><p>The blog didn’t display the complete article.</p><p>Instead, it showed only a portion of the content and eventually required a Chinese phone number to continue reading.</p><p>At first glance, this looked like a dead end.</p><p>However, there was an important clue:</p><p>Searching the page with the browser’s normal Ctrl + F showed that certain text existed somewhere on the page—even though it wasn&#39;t visible.</p><p>That immediately suggested that the content might still exist in the <strong>DOM</strong>.</p><h3>Solution</h3><p>Open the browser’s Developer Tools:</p><p><strong>Right-click → Inspect</strong></p><p>Then search within the HTML/DOM using:</p><pre>Ctrl + F</pre><p>and search for:</p><pre>.onion</pre><p>The hidden section of the article appeared inside the developer tools.</p><p>This revealed the Tor address.</p><h3>OSINT Lesson</h3><p>Never assume that something you can’t visually see doesn’t exist.</p><p>Modern websites frequently:</p><ul><li>dynamically load content,</li><li>hide elements with CSS,</li><li>render content through JavaScript,</li><li>truncate articles,</li><li>or display only part of the DOM.</li></ul><p>Developer Tools can therefore become an important OSINT resource.</p><h3>Challenge 2 — Pivoting from Tor to Telegram</h3><h3>Objective</h3><p>Identify the clearnet Telegram channel associated with the .onion site.</p><p>Finding a Tor domain is useful, but the next question is:</p><p><strong>Does the organization have a presence outside the Tor network?</strong></p><h3>Solution</h3><p>Take the .onion domain discovered in Challenge 1 and search for it using Google.</p><p>Putting the domain inside quotation marks makes the search more targeted:</p><pre>&quot;example.onion&quot;</pre><p>Search results can reveal references to the domain on:</p><ul><li>Telegram tracking websites,</li><li>forums,</li><li>cached pages,</li><li>directories,</li><li>social-media platforms.</li></ul><p>In this case, the search led to references associated with a Telegram channel.</p><p>The channel name was also interesting because it resembled the first few characters of the unique .onion domain.</p><p>That provided additional confidence that the two identities were connected.</p><h3>OSINT Lesson</h3><p>A Tor presence doesn’t necessarily mean an entity is exclusively on Tor.</p><p>Always test for <strong>clearnet ↔ darknet pivots</strong>.</p><p>A unique domain is particularly valuable because it can function as a fingerprint across search engines and third-party indexing services.</p><h3>Challenge 3 — Finding the Telegram Promoter’s QQ ID</h3><h3>Objective</h3><p>Identify the QQ messenger ID of an individual advertising gambling platforms through the Telegram channel.</p><h3>The Problem</h3><p>The channel contained thousands of messages.</p><p>Manually scrolling through everything would be extremely inefficient.</p><h3>Solution</h3><p>Use Telegram’s built-in search functionality.</p><p>The challenge provided several marketing-related keywords.</p><p>Search for one of those terms inside the channel.</p><p>This narrowed the results down to the relevant promotional post.</p><p>At the bottom of the post, the promoter had included a <strong>QQ ID</strong>.</p><p>That QQ number became the next pivot.</p><h3>OSINT Lesson</h3><p>When investigating large Telegram channels, don’t scroll.</p><p><strong>Search.</strong></p><p>Look for:</p><ul><li>usernames,</li><li>phone numbers,</li><li>email addresses,</li><li>domains,</li><li>cryptocurrency addresses,</li><li>advertisements,</li><li>repeated phrases,</li><li>contact information.</li></ul><p>A single unique keyword can reduce thousands of messages to a handful of relevant results.</p><h3>Challenge 4 — Documenting the Telegram Identity</h3><h3>Objective</h3><p>Record the promoter’s:</p><ul><li>official Telegram username</li><li>Chinese biography</li></ul><h3>Solution</h3><p>Click the sender’s name directly from the promotional post.</p><p>This opens the user’s Telegram profile.</p><p>Record the information exactly as presented.</p><p>For OSINT investigations, accuracy matters.</p><p>For example, don’t normalize or “correct” a username just because it looks unusual.</p><p>Record:</p><pre>Username:<br>Bio:<br>QQ:<br>Numeric Telegram ID:</pre><h3>Important Tip</h3><p>Usernames can potentially be changed.</p><p>A platform’s <strong>numeric account identifier</strong>, where available, is generally more useful for identity correlation than a mutable username.</p><p>This is a good general principle:</p><blockquote><strong><em>Prefer stable identifiers over mutable identifiers.</em></strong></blockquote><h3>Challenge 5 — Pivoting from QQ to a Phone Number</h3><p>Now the investigation becomes more interesting.</p><h3>Objective</h3><p>Use the QQ ID to identify associated telephone numbers.</p><h3>The Problem</h3><p>Normal Google searches didn’t directly reveal the database required for the lookup.</p><p>The challenge provided a clue:</p><p>The endpoint followed a naming pattern similar to:</p><pre>&lt;platform&gt;_tell.html</pre><p>and the website began with:</p><pre>av...</pre><h3>Solution</h3><p>Use Google Dorking.</p><p>Search for:</p><pre>inurl:q_tell.html</pre><p>The inurl: operator tells Google to prioritize pages containing a specific string within their URL.</p><p>The search revealed the relevant Chinese reverse-lookup database.</p><p>Enter the QQ ID into the appropriate lookup page.</p><p>The result returned <strong>two associated telephone numbers</strong>.</p><h3>OSINT Lesson</h3><p>Google Dorking isn’t only useful for finding exposed files or vulnerable infrastructure.</p><p>It can also be used for <strong>discovering obscure web applications and forgotten search endpoints</strong>.</p><p>Useful operators include:</p><pre>inurl:<br>intitle:<br>site:<br>filetype:</pre><p>The important skill isn’t memorizing hundreds of dorks.</p><p>It’s recognizing <strong>URL and application naming patterns</strong>.</p><h3>Challenge 6 — Tracing the Corporate Infrastructure</h3><h3>Objective</h3><p>Identify the corporate website/domain associated with one of the phone numbers.</p><p>The first phone number ended in:</p><pre>958</pre><h3>Solution</h3><p>Search the complete number in quotation marks:</p><pre>&quot;XXXXXXXX958&quot;</pre><p>The search results revealed a corporate domain associated with the number.</p><p>However, there was a problem.</p><p>The website was no longer operational.</p><p>Opening it produced a connection error.</p><p>At this point, many investigators might assume the lead is dead.</p><p>It wasn’t.</p><h3>Challenge 7 — Bringing the Dead Website Back to Life</h3><h3>Objective</h3><p>Recover the company’s official English/Pinyin name.</p><h3>Tool</h3><p><strong>Wayback Machine</strong></p><p>The Internet Archive’s Wayback Machine is one of the most valuable tools in historical OSINT.</p><p>Instead of asking:</p><blockquote><em>“What does the website look like now?”</em></blockquote><p>we can ask:</p><blockquote><em>“What did this website look like in the past?”</em></blockquote><h3>Solution</h3><p>Copy the inactive domain into the Wayback Machine.</p><p>Look through the available snapshots.</p><p>A snapshot from <strong>2025</strong> contained a functioning version of the corporate website.</p><p>Opening the archived page revealed the company’s information.</p><p>The page could then be translated to identify the official English/Pinyin company name.</p><h3>OSINT Lesson</h3><p>A dead domain isn’t necessarily a dead lead.</p><p>Historical infrastructure can contain:</p><ul><li>company names,</li><li>employee names,</li><li>contact numbers,</li><li>QQ accounts,</li><li>email addresses,</li><li>addresses,</li><li>product information,</li><li>social-media links.</li></ul><p>This is why <strong>historical web analysis</strong> is such an important part of OSINT.</p><h3>Challenge 8 — Finding a Secondary QQ Account</h3><p>While examining the archived corporate website, another useful piece of information appeared.</p><h3>Objective</h3><p>Find a second QQ number associated with the company.</p><h3>Solution</h3><p>Return to the archived page discovered in Challenge 7.</p><p>Check areas such as:</p><ul><li>header,</li><li>footer,</li><li>contact page,</li><li>customer support section,</li><li>administrative contact information.</li></ul><p>The archived website contained another QQ contact.</p><p>This secondary QQ ID became our next pivot.</p><h3>OSINT Lesson</h3><p>Don’t stop when you find the answer to a challenge.</p><p>Look around the surrounding evidence.</p><p>A page containing one identifier often contains <strong>additional identifiers</strong> that can lead to completely new branches of an investigation.</p><h3>Challenge 9 — Secondary QQ → Phone Number</h3><p>We now had a second QQ ID.</p><h3>Objective</h3><p>Identify the telephone number associated with it.</p><h3>Solution</h3><p>Return to the reverse-lookup database discovered during Challenge 5.</p><p>Use the secondary QQ number.</p><p>The database returned another associated phone number.</p><h3>Challenge 10 — Determining the Phone Number’s Origin</h3><h3>Objective</h3><p>Identify:</p><ul><li>telecommunications carrier</li><li>original registration region/city</li></ul><p>for the second phone number.</p><h3>The Problem</h3><p>General-purpose AI tools and maps aren’t always reliable for Chinese telephone-number attribution.</p><p>Carrier and registration-region information can depend on specialized local databases.</p><h3>Solution</h3><p>Use a dedicated Chinese mobile-number lookup service.</p><p>Enter the phone number and query the database.</p><p>The result provided information about:</p><pre>Carrier: China Mobile<br>Region: [Registered Chinese region]</pre><h3>OSINT Lesson</h3><p><strong>Use the right database for the geography.</strong></p><p>A common OSINT mistake is assuming that a globally popular tool is automatically the best source.</p><p>For region-specific investigations, local search engines and specialized databases can be significantly more useful.</p><h3>Challenge 11 — Phone Number → Weibo</h3><p>The investigation now moved from QQ and corporate infrastructure to Chinese social media.</p><h3>Objective</h3><p>Identify the target’s Weibo account using the phone number.</p><h3>Solution</h3><p>The same reverse-lookup database contained another useful endpoint.</p><p>The relevant endpoint followed the naming pattern:</p><pre>tell_wb.html</pre><p>where:</p><pre>tell → telephone<br>wb → Weibo</pre><p>Using the target’s phone number returned a <strong>numeric Weibo User ID</strong>.</p><p>We now had another pivot:</p><pre>Phone Number<br>     ↓<br>Weibo User ID</pre><h3>Challenge 12 — Getting Around the Weibo Login Wall</h3><h3>Objective</h3><p>Find the target’s Weibo username and biography.</p><h3>The Problem</h3><p>Opening the standard desktop profile URL:</p><pre>weibo.com/u/[USER-ID]</pre><p>resulted in a login requirement.</p><p>The information we needed wasn’t easily accessible through the standard desktop interface.</p><h3>Solution</h3><p>Try Weibo’s mobile web interface:</p><pre>m.weibo.cn/u/[USER-ID]</pre><p>The mobile interface exposed the profile information without requiring the same desktop login flow.</p><p>This allowed the investigation to recover:</p><ul><li>username</li><li>biography</li><li>historical posts</li><li>post metadata</li></ul><h3>OSINT Lesson</h3><p>Different interfaces of the same platform can expose different amounts of information.</p><p>When a desktop interface blocks access, it can be worth testing:</p><pre>Mobile web<br>Old URLs<br>Alternate subdomains<br>Public API endpoints<br>Cached versions<br>Search-engine indexes</pre><p>Always stay within the platform’s terms and the authorization boundaries of your investigation.</p><h3>Challenge 13 — Extracting Historical Device Metadata</h3><p>And finally, we reached the last challenge.</p><h3>Objective</h3><p>Determine the specific mobile device used by the target.</p><p>At first glance, recent posts only showed a generic:</p><pre>iPhone Client</pre><p>That wasn’t specific enough.</p><h3>Solution</h3><p>Instead of focusing only on recent posts, move backward through the user’s historical posts.</p><p>Older posts contained more detailed client/device information.</p><p>Eventually, one of the historical posts revealed the specific device:</p><p><strong>iPhone 7 Plus</strong></p><p>This completed the final pivot.</p><h3>The Complete Investigation Chain</h3><p>The entire investigation can now be represented as:</p><pre>                     ┌──────────────────┐<br>                     │   Chinese Blog   │<br>                     └────────┬─────────┘<br>                              │<br>                       Hidden .onion<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │  Tor / .onion    │<br>                     └────────┬─────────┘<br>                              │<br>                         Google Search<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │ Telegram Channel │<br>                     └────────┬─────────┘<br>                              │<br>                       Promotional Post<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │      QQ ID       │<br>                     └────────┬─────────┘<br>                              │<br>                      Reverse Lookup<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │   Phone Number   │<br>                     └────────┬─────────┘<br>                              │<br>                       Google Search<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │ Corporate Domain │<br>                     └────────┬─────────┘<br>                              │<br>                       Wayback Machine<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │ Archived Website │<br>                     └────────┬─────────┘<br>                              │<br>                       Secondary QQ<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │   Phone Number   │<br>                     └────────┬─────────┘<br>                              │<br>                     Mobile Lookup DB<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │ Carrier / Region │<br>                     └────────┬─────────┘<br>                              │<br>                       Phone → Weibo<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │   Weibo User ID  │<br>                     └────────┬─────────┘<br>                              │<br>                       Mobile Weibo<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │ Weibo Profile    │<br>                     └────────┬─────────┘<br>                              │<br>                      Historical Posts<br>                              │<br>                              ▼<br>                     ┌──────────────────┐<br>                     │   iPhone 7 Plus  │<br>                     └──────────────────┘</pre><h3>OSINT Tools Used</h3><p><strong>Browser Developer Tools</strong><br> Used to inspect the webpage DOM and locate the hidden .onion link.</p><p><strong>Google Search</strong><br> Used for general web discovery and correlating domains, phone numbers, and corporate infrastructure.</p><p><strong>Google Dorking</strong><br> Used to discover specific reverse-lookup endpoints using URL-based search operators.</p><p><strong>Telegram</strong><br> Used to locate the relevant promotional posts and identify the associated QQ account.</p><p><strong>QQ</strong><br> Used as an identity pivot during the investigation.</p><p><strong>Chinese Reverse Lookup Database</strong><br> Used for QQ → phone number and phone number → Weibo pivots.</p><p><strong>Wayback Machine</strong><br> Used to recover information from the dead corporate website.</p><p><strong>Web Archives Browser Extension</strong><br> Used to quickly check multiple web-archive sources.</p><p><strong>Reverse Image Search Extensions</strong><br> Useful for cross-platform image verification.</p><p><strong>Chinese Mobile Number Lookup</strong><br> Used to identify the carrier and original registration region of the phone number.</p><p><strong>Weibo</strong><br> Used to identify the profile and investigate historical posts.</p><p><strong>Mobile Weibo (</strong><strong>m.weibo.cn)</strong><br> Used to access public profile information when the desktop interface required login.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e7c46f61c76" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[UK OSINT CTF]]></title>
            <link>https://medium.com/@trivediaarsh51/uk-osint-ctf-342cf76e6202?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/342cf76e6202</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Mon, 06 Jul 2026 09:48:44 GMT</pubDate>
            <atom:updated>2026-07-06T09:48:44.453Z</atom:updated>
            <content:encoded><![CDATA[<h3>CTF Walkthrough</h3><h3>1. Bypassing the Gate :</h3><p>The challenge provided a CSDN blog post that appeared to be locked behind a login requirement, preventing access to the full article. Since the prompt hinted that the hidden content could still be accessed, the first step was to verify whether the data was actually missing or simply concealed.</p><p>Searching the page with <strong>Ctrl + F</strong> for <strong>.onion</strong> returned multiple matches, even though no .onion address was visible in the article. This indicated that the content was present somewhere on the page but hidden from the user interface.</p><p>To investigate further, I opened <strong>Developer Tools</strong> (F12) and used <strong>Inspect Element</strong>. Inside the <strong>Elements</strong> panel, I searched again for <strong>.onion</strong>. This revealed the hidden portion of the article in the page&#39;s <strong>DOM (Document Object Model)</strong>, allowing me to extract the complete .onion address.</p><p>This works because <strong>Inspect Element</strong> displays the <strong>live DOM</strong> — the webpage after the browser has processed the HTML, CSS, and JavaScript. Many modern websites load the complete content into the browser and simply hide it using frontend techniques such as CSS or overlays. Since the data already exists in the DOM, it can often be viewed through the browser’s developer tools.</p><p>In contrast, <strong>View Page Source</strong> only shows the <strong>original HTML</strong> sent by the server before any JavaScript executes. If content is added or modified dynamically after the page loads, it may not appear in the page source, whereas it will be visible in the live DOM shown by Inspect Element.</p><p>After locating the hidden text, I copied the .onion address and submitted it in the required format:</p><pre>osintuk{cabyceogpsji73sske5nvo45mdrkbz4m3qd3iommf3zaaa6izg3j2cqd.onion}</pre><p><strong>Key Takeaway:</strong> Frontend restrictions are not equivalent to access control. If sensitive data is sent to the client, inspecting the live DOM can often reveal information that is hidden from the rendered page.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=342cf76e6202" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[THM — WriteUp — 0x41haz]]></title>
            <link>https://medium.com/@trivediaarsh51/thm-writeup-0x41haz-afa8433598d9?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/afa8433598d9</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Tue, 03 Feb 2026 06:24:32 GMT</pubDate>
            <atom:updated>2026-02-03T06:24:32.730Z</atom:updated>
            <content:encoded><![CDATA[<h3>THM — WriteUp — 0x41haz</h3><p>Our goal is to locate a password that has been concealed within this particular file, according to the instructions concerning this challenge.</p><p>Moreover, it appears as if this program has many anti-reverse engineering mechanisms built into it that will make finding the Flag more difficult.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*rOzMcB-o-o0K4M0DmdhAhA.png" /></figure><p>Download the File &amp; start doing a static analysis</p><p>I start using the command : <strong>file</strong></p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*THI0h9tJ6ETCPZldjaN84Q.png" /></figure><p>Learn More about ELF :</p><p><a href="https://medium.com/@ajmewal/basics-of-elf-executable-and-linkable-format-file-88a516877356">Basics of ELF (Executable and Linkable Format) file</a></p><ul><li>Upon finding this problem I saw that the header contained the wrong information with regard to the endianness.</li><li>It seems that the ELF file is being imported in big-endian format instead of the indicated little-endian format which leads to tools being unable to read the ELF file correctly.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/955/1*A6tFz52RrlWVL2o2-FcNLw.png" /></figure><p>Edit the file header by the <strong>Hexeditor, </strong>convert the Big Endian to the Low Endian.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*CtabDthd45o1BiaaWOk8Vw.png" /></figure><p>Learn more<strong> Radare2 :</strong></p><p><a href="https://www.kali.org/tools/radare2/">radare2 | Kali Linux Tools</a></p><p>After Modifying , use r2 see the pseudocode of the File. To open it :</p><ul><li>r2 hex_modify</li><li>s main</li><li>VV</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/605/1*Yn2bxxhu6kBnoGuDpHXPOw.png" /></figure><ul><li>Merge all the Strings to form the Flag.</li><li>Verify it by submitting into the File.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/271/1*IdyDzFo-yYLWQuQCVfDbBw.png" /></figure><ul><li>Lastly, Wrap it into the Flag format of THM.</li></ul><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=afa8433598d9" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Deployment Architecture: How Who_Fi Actually Runs on Real Wi‑Fi Networks]]></title>
            <link>https://medium.com/@trivediaarsh51/deployment-architecture-how-who-fi-actually-runs-on-real-wi-fi-networks-4a8f04bc1101?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/4a8f04bc1101</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Tue, 09 Dec 2025 12:20:07 GMT</pubDate>
            <atom:updated>2026-07-28T09:42:39.768Z</atom:updated>
            <content:encoded><![CDATA[<p>From lab prototype to home and MDU rollouts using the routers you already own</p><ul><li>Who_Fi takes Wi‑Fi sensing from “cool research demo” to something that can actually live inside the same routers and mesh systems people already use at home or in apartments. Instead of adding new cameras or wearables, it reads subtle changes in Wi‑Fi Channel State Information (CSI) to recognize and re‑identify people as they move through a space.​</li><li>In this part of the series, the focus shifts from what Who_Fi is to where it actually runs: on the home gateway or access points, on an edge box, and in the cloud services that manage models, identities, and alerts. Readers should walk away with a mental picture of “router ↔ edge ↔ cloud,” how CSI flows between those layers, and which pieces are needed for a real deployment rather than a lab setup.​</li><li>The scope is intentionally narrow and practical: single‑family homes with a single router or mesh system, and multi‑dwelling units (MDUs) with managed Wi‑Fi across many apartments. That means every pattern and diagram you see here is something an ISP, building operator, or power user could realistically deploy on top of commodity Wi‑Fi 6/6E gear, not exotic hardware.​</li></ul><h3>Section 2: Building Blocks &amp; Roles</h3><ul><li>At the heart of a Who_Fi deployment are a few simple characters: your Wi‑Fi access points or gateways (where CSI is captured), an edge or cloud service that runs the deep learning models, and a management layer that stores identities, policies, and events. Think of it as “router with extra senses” at the bottom, “Who_Fi brain” in the middle, and “dashboards and integrations” at the top, all wired together over the same network links you already use.​</li><li>On the device side, compatible Wi‑Fi chipsets expose Channel State Information so the AP or gateway can sample how signals change when someone moves, then run lightweight pre‑processing (filtering, normalization, maybe simple feature extraction) before shipping that data or embeddings upstream. This keeps raw radio data close to the metal while giving the Who_Fi pipeline a clean, structured stream it can turn into biometric signatures.​</li><li>In the Who_Fi “brain,” a modular deep learning stack turns CSI sequences into fixed‑length vectors that act like radio fingerprints for each person, then compares those against a gallery of known signatures to decide “who just walked past the router.” Around that core, there is glue logic for identity lifecycle (enrol, update, revoke), event generation (enter/exit, anomaly), and integration with apps like smart‑home platforms, building management systems, or security dashboards.​</li><li>Finally, the control and analytics layer gives operators and power users tools to configure sensing zones, tune sensitivity, and monitor performance across many homes or apartments from a single pane of glass. In an MDU, this looks very similar to a managed Wi‑Fi setup today — just with extra tiles and graphs for presence, movement, and identification instead of only throughput and signal strength.​</li></ul><blockquote>“Any sufficiently advanced technology is indistinguishable from magic.” — Arthur C. Clarke</blockquote><blockquote>That is exactly the vibe here: from the outside it looks like a normal Wi‑Fi deployment; under the hood, the network quietly learns who is moving through the airwaves.</blockquote><h3>Section 3: Deployment Flow</h3><p>Once the hardware and software blocks are in place, the next question is simple: how does Who_Fi actually come alive in a real home or apartment network? The usual flow starts with installation on a Wi‑Fi gateway or access point, where the system begins capturing CSI from normal wireless traffic and then turns that raw signal into something the model can understand.</p><p>First comes initial setup and calibration. The system learns the room, the RF environment, and the placement of the access points, because Wi‑Fi sensing is highly affected by layout, wall material, furniture, and device placement. Think of this as the “teach the router the room” stage: the deployment is technically live, but it still needs a short adaptation period before it becomes useful.openaccess.</p><p>Next is enrollment. Known users are added to the identity gallery so the model can later match incoming CSI patterns against a set of trained signatures and say, “this is someone we already know.” In a practical setup, this is where the system links sensed signatures to labels, policies, or access rules depending on the use case.</p><p>After that, the system moves into continuous inference. As people move through the sensing zone, the gateway or edge stack keeps sampling, pre‑processing, and feeding CSI into the model, which then outputs identity matches, presence events, or anomaly signals in near real time. This is the stage where Who_Fi stops being a demo and starts behaving like a live service.</p><p>Because real environments never stay still, the deployment also needs ongoing adaptation. New furniture, changing traffic patterns, added devices, and seasonal RF changes can all shift performance, so the system should support periodic recalibration, model updates, and drift monitoring. In other words, the network should keep learning instead of assuming the apartment looked exactly the same on day 1 and day 101</p><p>Finally, a good deployment flow includes operator controls and monitoring. That means dashboards for sensing health, sensitivity tuning, event logs, and update management so that the system can be maintained across one home or many apartments without constant manual intervention. The cleanest way to think about it is: install, calibrate, enroll, run, adapt, and monitor — then repeat as the environment changes.</p><figure><img alt="This is an Image of the Flow of the deployment" src="https://cdn-images-1.medium.com/max/1024/1*aohXFAvb0TEv5m-aTgiymg.png" /></figure><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4a8f04bc1101" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Your Wi-Fi Router Can See You: Meet WhoFi, the Future of Invisible Identification]]></title>
            <link>https://medium.com/@trivediaarsh51/your-wi-fi-router-can-see-you-meet-whofi-the-future-of-invisible-identification-b3acd0eb5cbd?source=rss-607e7db4451b------2</link>
            <guid isPermaLink="false">https://medium.com/p/b3acd0eb5cbd</guid>
            <dc:creator><![CDATA[Trivedi Aarsh]]></dc:creator>
            <pubDate>Tue, 25 Nov 2025 14:19:15 GMT</pubDate>
            <atom:updated>2025-12-09T09:24:46.020Z</atom:updated>
            <content:encoded><![CDATA[<h3>Your Wi-Fi router can see you. Meet WhoFi. It is the future of invisible identification</h3><h4>Radio waves in your home can recognize who you are. They do this without cameras. No passwords are needed. No gadgets either.</h4><h3>A door opens without a key.</h3><p>Picture walking into your smart home. You have not even taken off your shoes yet. The lights adjust to your brightness preference right away. Your favorite playlist starts up. The thermostat moves to your ideal temperature. The security system notes your arrival.<br>Here is the magic.</p><ul><li>Your home did not use any camera.</li><li>No facial recognition happened.</li><li>No fingerprint scanner was involved.</li><li>No fancy wearable device came into play.</li></ul><p>It recognized you through the invisible radio waves that fill every room. Those are your Wi-Fi signals.</p><p>Welcome to <strong>WhoFi — </strong>This is a groundbreaking technology. It will change how we think about identification. Privacy and smart homes will never be the same.</p><h3>Why this Matters right now ?</h3><p>We live in a world obsessed with surveillance. Cameras watch us everywhere. They are in offices. They cover streets. They even invade homes. Facial recognition systems track our movements. What if there is a way to identify people that preserves privacy completely. It would be different.<br>This way is completely invisible. No cameras. No images. Just pure radio waves.</p><ul><li>It puts privacy first. No visual data gets captured.</li><li>It is already available. Your Wi-Fi router has the hardware.</li><li>It works anywhere. Through walls. In darkness. In any weather.</li><li>It achieved 95.5 percent identification accuracy in research.</li></ul><p>Does it sound like science fiction. It is not. This is the evolution of Wi-Fi technology. We never imagined it could go this far.<br>Let us demystify how it works.</p><h3>Section 1: The Invisible World Around You — Understanding CSI</h3><h3>The Lake Analogy (Really Understanding This)</h3><p>Picture yourself at a calm and peaceful lake. You pick up a rock and toss it into the water. When it hits, ripples will form on the surface of the water, moving out in concentric circles, away from where you threw it, hitting all the way around all the way back to where you are standing.</p><p>Now, if you were able to accurately measure these ripples as they return to where you are standing, you would be able to determine :</p><ul><li>what you threw</li><li>where it was thrown from</li><li>how much it weighed.</li></ul><p>CSI uses the same principals for Wi-Fi signals as you would for determining what kind of rock you threw and how much it weighed as it would return to you in the above example.</p><h3>What is CSI? The Fingerprint of Radio Waves</h3><p>CSI or Channel State Information is an indicator that can be measured according to how Wi-Fi signals are altered as they travel through many mediums: from a Transmitter, to the object, then to the receiver.</p><p>When radio waves strike an object, these radio waves produce an individual “signature” based upon that object. Each person creates an individual signature based upon the shape of their body, the bone density of their body, and the amount of water in their body. An individual’s body signature is created based upon the way their body interacts with radio waves will be as distinctive to a person’s fingerprint.</p><p>CSI is unique because:</p><p>Radio waves pass through walls and therefore can be utilized to determine body signatures from a distance, regardless of the obstacles involved.</p><p>Radio wave technology works without requiring any amount of light.</p><p>All of the internal structures within an individual interact with radio waves (bones, organs, etc.) and interact to create an encoded record of the person based upon their unique anatomy.</p><p>CSI consists of multiple levels of information.</p><ul><li>CSI = Amplitude (i.e., signal strength) + Phase (i.e., signal timing).</li></ul><p>Amplitude is the “loudness” of a person’s voice that is represented within the Wi-Fi signal.</p><ul><li>Phase is the “timing” or “rhythm” of that voice.</li></ul><p>When combined (amplitude and phase), these two data sets will provide the whole story of how an individual’s body uniquely changes radio waves.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*vgzlUBGhFA8dj697_Ru8Bg.png" /></figure><h3>Section 2: The Technology that Makes WhoFi Possible-Multiple Antenna and Multiple Frequency Modulation.</h3><h4>How Modern Wi-Fi is Perfect for Detecting Movement ?</h4><p>Modern Wi-Fi routers have two special technologies that make WhoFi possible.</p><h4>MIMO (Multiple Input, Multiple Output)</h4><p>Modern Wi-Fi has multiple antennas at the transmitter and multiple antennas at the receiver as opposed to just one antenna for each.</p><p>This is similar to having multiple security cameras situated around a person from different angles.</p><p>Why this is Important for WhoFi:</p><p>Since there are multiple antennas at both the transmitter and receiver, there are multiple paths for the transmission of signals through space.</p><p>Therefore, even if one of the paths is obstructed by a wall, the other antennas will still be able to detect the interaction of a person’s body with those signals. They also provide a three-dimensional picture of how a person affects the signals viewed from multiple angles.</p><h4>OFDM (Orthogonal Frequency-Division Multiplexing)</h4><p>Rather than transmitting every single piece of WiFi information at a single frequency, as done in conventional wireless technology, the manufacturer of WhoFi opted to transmit the data over many “sub-frequencies”, or “Subcarriers”. Through WhoFi’s research and experiments, there were found to be 114 subcarriers.</p><p>This information is important for WhoFi :</p><ul><li>Different types of frequencies will react with your body in many different ways; e.g. A low frequency will go through your skin and into the tissue.</li><li>A high frequency will bounce around the outside.</li><li>With 114 different “views”, it creates a very complex and complete fingerprint of your body, having 114 different images, similar to an X-ray taken at different wavelengths.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*uIQ7ECMKEWe0WWS5O0DfvQ.png" /></figure><h3>Section 3: The Data Deluge — What Is the Access Point’s Magnitude of Indoor Location?</h3><p>When a person makes a 2m walk past an access point with a WiFi Network setup, each 2m walk results in 684,000 total data points being generated by 3 RX Antennas, 114 Subcarriers, and 2000 Time Data Samples.</p><p><em>That’s not 684 measurements. That’s 684,000 measurements !</em></p><p>Each measurement contains information about how a specific radio wave was affected by your body because of how that radio wave is being measured at this particular point in time. Imagine if there were 6,84,000 small sensors located all around you in every direction and every frequency (in this case, the frequency bands are sub-carriers) taking measurements of how well your body was impacting that radio wave.</p><p>This is why WhoFi works. WhoFi utilizes the uniqueness of CSI Data associated with each individual (which means there are many millions of individuals) because there is so much individual information that it becomes impossible to determine who the individual was based on CSI measurements alone.</p><h3>Section 4: Your Body as a Radio Wave Fingerprint Adaptation</h3><p>Human bodies distort Wi-Fi signals differently than one another for several reasons:</p><h4>Why?</h4><ul><li>Different body shapes = unique scattering patterns</li><li>Bone structures differences = unique impedance to radio waves</li><li>Gait differences = different walking signatures</li><li>Different water content = unique electromagnetic properties of individuals</li></ul><p>Therefore, it can be said that each individual has a unique “voice” on the Wi-Fi spectrum.</p><p><strong>The following numbers were derived from the WhoFi research according to their experimentation using 14 different individuals.</strong></p><p>In experiments with just 14 subjects:</p><ul><li>LSTM encoder: 77.7% accuracy identifying people</li><li>Bi-LSTM encoder: 84.5% accuracy</li><li>Transformer encoder: 95.5% accuracy</li></ul><p>If the Transformer encoder correctly identifies individuals 95.5% of the time, that means the system accurately identifies individuals with nearly every single match at the top of the list</p><h3>Section 5: Raw channel state information (CSI) reports contain a lot of irrelevant information, including:</h3><p>It contains:</p><ul><li>Random noise caused by environment</li><li>Hardware imperfections causing noise</li><li>Noise from other Wi-Fi networks</li><li>Reflections of signals that are not relevant to the user or application</li></ul><p>So, in order to use the data for identification, we will first need to “clean” the data, which is called preprocessing.</p><h4>To clean data, we use :</h4><h4>Step 1: Amplitude Filtering (Removing the Spikes)</h4><p>To filter out the noise in the amplitude of the reported signal. That is, we utilize an amplitude filtering technique called the Hampel Filter to remove random spikes from the reported CSI data.</p><p>In our graph of the reported signal strength over time, you will mostly see smooth transitions in amplitude, but you will notice some occasional sharp spikes/quick drops.</p><p>By filtering out the random spikes/drops with the Hampel Filter, we will keep only those reported strength readings that were caused by the user’s body.</p><h4>Step 2: Phase Sanitization (Removing Hardware Bias)</h4><p>Removing the effects of any imperfections in the user’s Wi-Fi hardware (hardware bias).</p><p>Specifically, the raw phase data that was collected from the user’s Wi-Fi hardware will likely contain a linear drift resulting from imperfections such as misalignment between Wi-Fi transmitter and receiver, cable delays, etc.</p><p>To correct for this drift, we fit a linear trend line to the drift and subtract the trend line from the reported raw phase data. What remains will be the actual phase shift that is caused by the user’s body.</p><h3>Section 6: The Brain of WhoFi — Neural Encoders</h3><p>Our data is in clear to see, this capture represents over 684,000 unique data points representing exactly how someone’s body interacts with Wi-Fi signals.</p><p>How do we translate a person’s unique body data into a unique ID code ?</p><ul><li>Using Neural Networks to Encode Body Data — The Encoder Device An encoder takes a long sequence of body data and compresses it down into a short fixed-size vector (signature).</li></ul><h4>Why Should we use an Encoder?</h4><p>Automatic Feature Extraction For AI Provides Consistent Signatures For</p><ul><li>Different Walking Styles A Feather In Your Cap That Can Handle Variations</li><li>Due To Clothes, Moods Or Time Of Day Ability To Store Compact Codes For Easy Comparison Between Similar Individuals ALGOS Compared (three different encoder types):</li></ul><h4>LSTM: Left To Right Step By Step Reader</h4><p>This type of encoder reads a person’s CSI data one timestep at a time, from left to right. It stores the short term history in its “short term memory”. WhoFi Data :</p><ul><li>Accuracy Of 77.7%</li><li>Limitations — The memory “fades” by the time the network gets to the end, just like in writing a long sentence, you may have forgotten the first part before you reach the end.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OPwkPs95lPVEr4zYhFH7Wg.png" /></figure><h4>Bi-LSTM: Dual Reader In Both Directions</h4><p>In essence, two encoders being read at the same time, one reading from left to right and the other from right to left, encodes the entire context of the body’s walking pattern. WhoFi Data</p><ul><li>Accuracy Of 84.5%</li><li>The advantage of the Bi-LSTM encoder is that it sees the entire walking pattern, not just small “snippets”.</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*3VAwL0aSFLetloFNhGjxgQ.png" /></figure><h4>Transformer: The Global Genius</h4><p>Simultaneously evaluates every one of its 2000 time frames and determines through “self-attention” which portions are important in relation to one another.</p><p>Achieved an accuracy of 95.5% between its estimates and the accuracy of the WhoFi dataset. The Transformer model can view events with large time gaps between them, stores everything forever, processes everything at once.</p><h3>Part 7: The Magic Moment — Creating Your own Radio Fingerprint</h3><p>After obtaining 684,000 point values, each encoder will produce one 256-length numeric result (known as a vector) , which serves as your unique <strong>Wi-Fi biometric fingerprint</strong>.</p><h4>How is it Unique?</h4><p>Your unique Wi-Fi fingerprint is determined by training our neural network to do two things at once:</p><ol><li>Group together the same-person signatures, so that if you took two separate walks, they would have similar patterns.</li><li>Separate from each other all different individuals, thus making sure that your signature would be different than everyone else’s</li></ol><p>This training technique is called<strong> in-batch negative loss.</strong></p><ul><li>It works by creating a pair of two similar patterns of walking by the same individual and minimizing the distance between them, while creating the greatest separation possible between any other individual’s walk patterns.</li></ul><h3>Part 8: The Final Step — Identifying You</h3><p>You walk into your smart home and your CSI data is collected, Preprocessed and encoded into a signature vector.</p><p>Now what?</p><p>The system compares your signature to a known signature (stored on the device or server):</p><p>Your Signature vs Alice Signature → 95% similarity</p><p>Your Signature vs Bob Signature → 12% similarity</p><p>Your Signature vs Darth Signature → 8% similarity</p><p>Winner: Alice (95%)</p><p>This is Re-Identification (Re-ID): matching a probe to a gallery.</p><p>Using cosine similarity (a mathematical metric of the anglee between the two vectors), The system instantly knows: This is you.</p><h3>Section 9: Why WhoFi is Changing the Game in Terms of:</h3><p><strong>Privacy:</strong> WhoFi does not capture images, videos, or facial recognition data. The only thing that is collected is how radio waves travel and interact with objects using physics.</p><p><strong>Availability</strong>: WhoFi works everywhere, even if the light is out. Its ability to identify an object through walls and other obstructions means it’s possible to identify someone through a wall.</p><p><strong>Infrastructure</strong>: WhoFi uses existing hardware built into routers, so there are no new installations needed for WhoFi; it works on existing infrastructure.</p><p><strong>Security</strong>: WhoFi uses radio signatures as a method for authenticating users, making it significantly more difficult to create a fake account than it would be to take a selfie of someone.</p><p><strong>The Smart Home Revolution</strong>: Imagine having your doors automatically unlocked for you and not having them unlocked for strangers. Your office’s environment will adjust based on who is present in the office. The ability to monitor your health without ever stepping foot in a doctor’s office while providing maximum privacy.</p><h3>Section 10: Limitations of WhoFi Today :</h3><p>Currently, WhoFi has some limitations that will need to be addressed in the future. They include:</p><ol><li>Limited number of users within range — WhoFi works well when the number of users is 14 or less, but it is unclear how well it will function for 1,000 or more users.</li><li>Location-specific models — The model has been trained on a single room and cannot be readily applied to other locations.</li><li>Clothing type affects user signature — The same user can produce different signatures based on whether they are wearing baggy or fitted clothing.</li><li>Multiple users moving at the same time — When multiple users are in motion at the same time, the system has difficulty identifying each of them.</li><li>Changes to the environment can change user signatures — Rearranging the furniture can impact the signal patterns.</li></ol><p>By obtaining more data, using transfer learning techniques and creating adaptive models, solutions can be developed for these limitations.</p><h3>Section 11: The Future is Already Here</h3><ul><li><strong>Healthcare</strong>: To passively monitor and track health of seniors through Wi-Fi without using video cameras.</li><li><strong>Smart Offices</strong>: To provide seamless access to secure space with no keys or cards.</li><li><strong>Vehicles</strong>: Detecting driver drowsiness through Wi-Fi sensing</li><li><strong>Security</strong>: To use two-factor authentication (biometrics + physical activity) to prevent unauthorized access to sensitive personal information.</li><li><strong>Smart Homes</strong>: To become a “smart” room by tracking who enters and exits the room.</li></ul><p>This is only a small part of what you can create with the understanding of how your body communicates with WiFi.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b3acd0eb5cbd" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>