<?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 Chris Hernandez on Medium]]></title>
        <description><![CDATA[Stories by Chris Hernandez on Medium]]></description>
        <link>https://medium.com/@piffd0s?source=rss-796ac520f43c------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*44niCUv973us39qT1fCyXg.jpeg</url>
            <title>Stories by Chris Hernandez on Medium</title>
            <link>https://medium.com/@piffd0s?source=rss-796ac520f43c------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Thu, 19 Nov 2026 14:32:27 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@piffd0s/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[When AI Finds a Bug, just not the bug you want]]></title>
            <link>https://piffd0s.medium.com/when-ai-finds-a-bug-just-not-the-bug-you-want-c109345a4598?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/c109345a4598</guid>
            <category><![CDATA[vulnerability]]></category>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[vulnerability-management]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Thu, 22 Oct 2026 20:44:41 GMT</pubDate>
            <atom:updated>2026-10-22T20:44:41.594Z</atom:updated>
            <content:encoded><![CDATA[<p><em>What PHP2Shell (and other bugs) have taught us about automated CVE analysis, unreliable public PoCs, and the continuing importance of hands-on vulnerability research</em></p><p>When the PHP2Shell vulnerabilities became public, our automated CVE-analysis system did exactly what it was designed to do. It collected advisories, searched for related code, examined public proof-of-concept repositories, and assembled a technically plausible explanation.</p><p>There was only one problem: the explanation was wrong and did not match the underlying root cause… without human eyes on the bug our AI was confident it had found the bug (it hadn’t). We had encountered this issue before on other CVEs and bug chains so in this case we were ready to dive deeper into the true root cause.</p><p>The bugs it identified were real or close enough to real to sound convincing. Some were in the right subsystem. Some caused crashes (oom). Some resembled the rapidly appearing PoCs on GitHub and elsewhere.</p><p>But they did not explain the original research. Details from the <a href="https://calif.io/php2shell/">Calif Research write-up</a> did not line up. Several proposed chains required application-level vulnerabilities or special configurations that were absent from the disclosure. They did not explain how exploitation could begin before application code ran, or why Apache mod_php, PHP-FPM, and other execution environments behaved differently.</p><p>Our system had found bugs. It had not found <em>the bugs</em>.</p><p>Worse, when we pushed it to achieve remote code execution, it began looking for ways to satisfy the requested outcome without proving the vulnerability.</p><p><strong>In plain language, the AI wants to cheat, and its very lazy.</strong></p><h3>An information environment full of traps</h3><p>PHP2Shell emerged under unusually difficult conditions for automated triage. The existence of two critical PHP vulnerabilities became public before the coordinated release. A demonstration video briefly appeared online. Screenshots, speculative mitigations, hastily assembled scanners, and purported exploits quickly followed.</p><p>The result was an information environment almost designed to mislead an automated system.</p><p>Models could retrieve hundreds of references to PHP sessions, upload progress, serialization, file inclusion, heap corruption, and RCE. What they could not determine from repetition alone was which sources had reproduced the actual vulnerability, which had discovered an adjacent primitive, and which were simply fabricating a convincing result.</p><p>Some public PoCs were mechanically invalid. They referenced code paths that could not produce the claimed effect or confused a crash with controlled memory corruption.</p><p>Others demonstrated real behavior but depended on an additional vulnerability, such as a file-inclusion sink or an attacker-controlled deserialization gadget. That might form an interesting exploit chain, but it did not prove PHP2Shell.</p><p>Some appeared crafted around the expected outcome: vulnerable helper endpoints, weakened configurations, prearranged application behavior, or other conditions that quietly supplied the missing step. A command ran at the end, but the claimed CVE was not necessarily what made it run.</p><p>And some were simply fake… Here is a list of the inaccurate POCs or exploits that don’t reach the vulnerable code paths. The risk being other automated systems ingest these and write wrong on inaccurate detections for vulnerabilities that cannot realistically be exploited.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*gfGAGT2slTtnddDgVR7FaA.png" /><figcaption>Incorrect php2shell exploit examples</figcaption></figure><p>This is not new in vulnerability research. Popular CVEs have long attracted mislabeled, nonfunctional, and occasionally malicious PoCs. What AI changes is the speed and volume at which <em>plausible-looking explanations</em> and exploit code can be produced, copied, summarized, and fed back into other models.</p><p>Once enough repositories repeat the same mistake, retrieval begins to mistake repetition for corroboration.</p><h3>Our agent tried to optimize for the wrong result</h3><p>We encountered the same problem inside our own system.</p><p>When we asked the agent to investigate the vulnerability, it helped identify promising source areas and candidate lifetime problems. But when we changed the objective to “achieve RCE” its behavior changed.</p><p>The agent began optimizing for the visible success condition: make a command execute.</p><p>It proposed paths that introduced assumptions not present in the original target. It was willing to rely on an additional application flaw, modify the environment, weaken a mitigation, or otherwise supply a convenient bridge between the memory bug and code execution.</p><p>Each shortcut could produce something that looked like success. None necessarily proved that the disclosed vulnerability caused it.</p><p>The agent was not being deceptive in a human sense. It was following an underspecified objective. <strong>We had asked for RCE, and it searched for the easiest route to RCE.</strong></p><p>But our actual research question was narrower and more specific:</p><blockquote><em>Can we demonstrate the claimed impact through the disclosed vulnerability, under the conditions described by the original researchers, without introducing another bug or changing the target to make the exploit work?</em></blockquote><p>Those are very different tasks.</p><p>That experience forced us to define what counted as proof before allowing the agent to continue:</p><ul><li>The result had to follow from the suspected root cause.</li><li>The target could not contain helper functionality added for the PoC. (AI Loves to do this)</li><li>Security properties such as ASLR could not be silently disabled.</li><li>Any additional application-level prerequisite had to be identified explicitly.</li><li>A memory-safety violation had to be confirmed independently, preferably with instrumentation such as AddressSanitizer.</li><li>The proposed chain had to explain the affected versions, execution environments, and application independence described in the disclosure.</li><li>A patched build had to serve as a negative control.</li></ul><p>Once those constraints became part of the task, the agent stopped trying merely to win and became useful for research again.</p><h3>A plausible answer is not a root cause</h3><p>Modern AI systems are exceptionally good at gathering and connecting information. Give an agent a CVE identifier and it can rapidly locate advisories, patches, commits, historical bugs, public exploits, and nearby source code.</p><p>During a fast-moving disclosure, however, the available information is neither equally reliable nor mutually consistent.</p><p>An automated system can easily find the densest explanation rather than the correct one. It can produce a coherent narrative in which every individual step sounds reasonable, even though the complete chain cannot account for the observed behavior.</p><p>That appears to be what happened during our initial PHP2Shell triage. The system anchored on familiar techniques involving session files, upload progress, serialization mismatches, and application-level file inclusion.</p><p>Those techniques were related to the attack surface, but relevance is not causality.</p><p>Reproducing a crash is not the same as reproducing a vulnerability. Demonstrating code execution through an additional application flaw is not the same as application-independent RCE. Finding a suspicious pointer in the patched subsystem is not the same as proving that it caused the disclosed impact.</p><p>The missing ingredient was not more search. It was deeper research and deliberate falsification.</p><h3>Following the contradictions</h3><p>We changed our approach. Instead of gathering more evidence that supported the leading theory, we started with the details that contradicted it.</p><p>The official advisory describes <a href="https://github.com/php/php-src/security/advisories/GHSA-w8hw-w9fx-j79w">CVE-2026–109618</a> as a problem in PHP’s handling of out-of-memory conditions. PHP used longjmp() to escape from an allocation failure. If that occurred while internal request data was only partially initialized, ordinary control flow was abandoned and request shutdown could later treat incomplete state as valid.</p><p>The advisory for <a href="https://github.com/php/php-src/security/advisories/GHSA-6wmf-4rv5-p3rp">CVE-2026–109619</a> describes the related lifecycle failure. Request-owned state could remain reachable through a worker-global pointer after the request’s memory had been released. A subsequent request handled by the same worker could consume that stale state before normal initialization reset it.</p><p>Reduced to its essential form, the pattern looked like this:</p><ol><li>PHP publishes internal state belonging to the current request.</li><li>A fallible operation triggers an OOM or another fatal interruption.</li><li>The corresponding initialization, rollback, or reset does not complete.</li><li>The request heap is released, but the worker survives.</li><li>A later request consumes the abandoned state as though it were still valid.</li></ol><p>Session upload progress was significant, but it was not the complete root cause. It was one mechanism capable of publishing or consuming session state before normal request initialization. The broader vulnerability involved interruption safety and the lifetime of state crossing request boundaries.</p><p>Once we understood that pattern, the disclosure began to make sense. It explained the affected-version range, why some paths ran before application code, why worker reuse mattered, and why the complete patch addressed a family of lifecycle failures rather than one narrowly defined pointer bug.</p><h3>Deep research remains the difference</h3><p>It is tempting to describe the initial failure as hallucination, but that is too simple.</p><p>Most importantly, the researcher knows when the system is attempting to satisfy the requested outcome instead of answering the research question.</p><p>The PHP2Shell disclosures themselves demonstrate how productive human–AI collaboration can be. Calif describes using an AI “audit buddy” to identify a vulnerable ordering and generalize the pattern. <a href="https://www.wiz.io/blog/php2shell-php-rce">Wiz’s independent research</a> reports extensive use of frontier models during discovery, validation, and exploit development.</p><p>But in both cases, the models operated inside a process directed by researchers who understood the target, challenged intermediate results, built appropriate experiments, and refused to treat plausible output as proof.</p><h3>How it changes our workflows going forward.</h3><p>We do have what I would call an “exploit quality agent” that runs after a cve has been validated. This is used to rank the exploit quality and answer the question “Is this bug useful to an attacker, and therefore high importance to a defender?” During the exploit development phase we have also developed a quality control piece to validate that the exploit the system develops matches the technical description of the CVE without cheating, making assumptions, or leveraging application behavior not defined.</p><p>For me the next frontier would be our system correctly one-shotting a more complex CVE like this in a fully autonomous way.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=c109345a4598" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[RefluXFS: From Advisory to Root Shell in Minutes]]></title>
            <link>https://piffd0s.medium.com/refluxfs-from-advisory-to-root-shell-in-minutes-4e334131ce68?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/4e334131ce68</guid>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Fri, 24 Jul 2026 03:01:23 GMT</pubDate>
            <atom:updated>2026-07-24T03:01:23.017Z</atom:updated>
            <content:encoded><![CDATA[<p>When Qualys published CVE-2026–64600 on July 22, no proof-of-concept existed. Our autonomous pipeline read the advisory, built a vulnerable environment, wrote a working kernel exploit, and confirmed uid=0 all without a human touching a keyboard.</p><h3>The advisory, and nothing else</h3><p>On July 22, 2026, the Qualys Threat Research Unit published their advisory for CVE-2026–64600 — a race condition in the Linux kernel’s XFS copy-on-write path they dubbed RefluXFS. The vulnerability had been present in every mainline kernel since version 4.11, shipped in 2017. Nine years of silent exposure.</p><p>The advisory described the bug in precise terms: xfs_reflink_fill_cow_hole() drops the inode lock (ILOCK) at line 458 to allocate a transaction, re-checks the refcount at line 468 without re-reading the data fork mapping, and a concurrent writer can complete an entire CoW cycle in that window. A textbook TOCTOU. The fix — commit 2f4acd0 — had been merged six days earlier, but most production systems hadn&#39;t patched yet.</p><p>What the advisory did <em>not</em> include was a proof-of-concept. No exploit code. No reproduction steps beyond a high-level description. For most security teams, this means waiting — watching GitHub, monitoring oss-security, hoping someone publishes a PoC before an attacker does.</p><p>We didn’t wait.</p><h3>What the pipeline saw</h3><p>Our CVE analysis pipeline ingested CVE-2026–64600 the same way it processes every new vulnerability: autonomously. No analyst queued it. No one wrote a prompt. The pipeline read the NVD entry, the Qualys advisory, and the oss-security mailing list post, then made a series of decisions:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/903/1*cCdVJgIw480YHmSBuofIoQ.png" /><figcaption>our exploitation</figcaption></figure><p>The pipeline classified this as a local kernel privilege escalation, spun up a Fedora 40 container with a 512 MB XFS loopback filesystem (reflink enabled), verified the host kernel was in the affected range (6.17.0, well below the 6.18.39 fix), and handed off to the exploit development agent.</p><h3>Writing the exploit from the advisory alone…</h3><p>This is the part that matters. The exploit agent had no reference implementation. No GitHub repo to clone. No blog post with copy-pasteable shellcode. It had the advisory’s description of five kernel functions, a line-number pair (458 and 468), and the semantic understanding that a refcount checked after an ILOCK drop can go stale.</p><p>From that, it derived the full exploitation chain:</p><p>Step 1 — Clone the target file (a root-owned /etc/passwd) using cp --reflink=always. This shares physical blocks between the clone and the original, setting the refcount to 2.</p><p>Step 2 — Spawn 32 threads issuing concurrent O_DIRECT writes to the clone. This creates contention on the ILOCK, maximizing the chance that one thread drops the lock while another completes a full CoW cycle.</p><p>Step 3 — When the race hits, one thread’s stale re-check sees refcount=1 at the original physical block address and writes the payload in-place — directly to the block backing /etc/passwd.</p><p>Step 4 — Invalidate the page cache with posix_fadvise(POSIX_FADV_DONTNEED) so the block-level corruption becomes visible to userspace reads.</p><p>The payload is a single line: r::0:0:r:/root:/bin/sh. A passwordless root-equivalent user. After the race wins, su - r drops into uid=0.</p><h3>The moment it lands</h3><p>The first test run in the lab container:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/373/1*01KwJ1BSWVuSMfiV6o3Y8A.png" /><figcaption>winning the race</figcaption></figure><p>Attempt 1. Seventeen milliseconds. From an unprivileged user account to a root shell, on a kernel that ships as the default on every RHEL 8+, Fedora, CentOS Stream, and Amazon Linux installation with XFS — which is to say, nearly all of them.</p><p>The race is not fragile. Across every test run, the exploit won within the first 10 attempts, typically under one second. The 32-thread contention pattern makes the ILOCK drop window trivially hittable.</p><h3>Why this was hard without a PoC</h3><p>Kernel filesystem race conditions are among the most difficult vulnerabilities to exploit from a description alone. The developer must understand:</p><p>The allocator’s lock semantics. XFS drops the ILOCK to avoid transaction-reservation deadlocks. This is a deliberate design choice, not a bug in isolation, yet it becomes a bug only because the re-check after re-acquisition uses stale states.</p><p>The page cache invalidation. The race corrupts blocks at the storage layer, below the page cache. Without posix_fadvise(POSIX_FADV_DONTNEED), reads return stale cached data and the exploit appears to fail even when the blocks are already corrupted on disk.</p><p>The refcount precondition. The target file’s blocks must start at refcount=1. Pre-existing reflink clones (from admin backups, for instance) raise the count and prevent the race from producing a stale re-check result. The exploit must create its own clone to control the initial state.</p><p>Each of these details was extracted from the advisory text and the kernel source, not from an existing exploit. The pipeline’s exploit agent read the fixing commit (2f4acd0), identified the exact code path, and derived the exploitation constraints from first principles.</p><h3>What this means for defense</h3><p>The window between advisory and weaponization used to be measured in days or weeks. For CVE-2026–64600, it was minutes. That is not because this vulnerability was simple. It is because the process of reading an advisory, understanding a code path, building a test environment, and deriving an exploit can now be automated end-to-end.</p><p>Defenders should assume the same compression. When Qualys publishes an advisory for a nine-year-old kernel bug affecting 16.4 million systems, the clock is already running. The patch (2f4acd0) was merged on July 16. The advisory dropped July 22. Working exploits exist by July 23. That leaves a remediation window measured in hours, not the weeks that traditional patch cycles assume.</p><p>Patch. Update to kernel ≥ 6.18.39 (or LTS backports 6.12.96, 7.1.4). There is no workaround — the only fix is the kernel update.</p><p>Detect. Deploy the YARA rules from our package. Monitor /etc/passwd for unexpected uid=0 entries. Watch for anomalous FICLONE ioctl bursts followed by O_DIRECT write storms — the behavioral fingerprint of this race.</p><p>Scope. Inventory every host running XFS with reflink enabled. On RHEL 8+ and derivatives, that is the default filesystem configuration. If xfs_info / shows reflink=1, the host is in the blast radius.</p><h3>The pipeline, not the person</h3><p>No human wrote this exploit. No analyst read the advisory and sat down with vim and gdb. The entire chain — from NVD lookup to confirmed root shell — was orchestrated by VulnQL’s autonomous CVE pipeline: a sequence of specialized agents that research, build, detect, exploit, and triage vulnerabilities at machine speed.</p><p>That is the capability we are building. We are closing the gap between disclosure and understanding before attackers close it first.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=4e334131ce68" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Running iOS in a Virtual Machine on Apple Silicon]]></title>
            <link>https://piffd0s.medium.com/running-ios-in-a-virtual-machine-on-apple-silicon-e8eab0d95798?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/e8eab0d95798</guid>
            <category><![CDATA[ios]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[iphone]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Wed, 15 Jul 2026 18:03:14 GMT</pubDate>
            <atom:updated>2026-07-15T18:03:14.207Z</atom:updated>
            <content:encoded><![CDATA[<p>There is a question that every mobile security team, enterprise MDM vendor, and every iOS CI/CD pipeline eventually runs into: How do you run iOS without dedicated and jailbroken hardware?</p><p>The iPhone is one of the most important computing platforms on Earth, but its software remains uniquely difficult to virtualize. Android has had robust emulators since its first SDK. Windows, Linux, and macOS all run happily inside hypervisors. iOS stands alone, locked behind custom silicon, a proprietary boot chain, and a kernel that actively resists running anywhere except on the device Apple intended.</p><p>To solve this problem we’re builing a native iOS hypervisor that boots a real XNU kernel on Apple Silicon hardware, using Apple’s own Hypervisor framework to run iOS at near-native speed.</p><p>The iOS Virtualization Gap</p><p>The commercial landscape for iOS virtualization has historically been narrow. A handful of companies have attempted cloud-based iOS device farms, typically built on jailbroken hardware or ARM server racks running very customized firmware. These solutions exist, but they share fundamental limitations:</p><ul><li>Hardware dependence. You need physical iPhones or iPads, hundreds of them, racked and managed, to offer any kind of scale.</li><li>Version inflexibility. Pinning a device to a specific iOS build means dedicating that hardware. Testing across a matrix of versions means a significant hardware investment… hardware that can be hard to aquire if you are interested in older (vulnerable) ios versions.</li><li>Cost at scale. At hundreds per version locked device, building a 50-device test lab is a six-figure capital expenditure before you write a line of test code or develop a harness.</li></ul><p>For security researchers, this gap is painful. Analyzing an iOS kernel vulnerability typically requires: a device running the exact vulnerable build, a way to attach a debugger, and the ability to snapshot and restore state. Meeting all three conditions simultaneously has historically required either extraordinary effort or access to resources most organizations don’t have.</p><p>Our Approach: Native Hypervisor, Real (hacked) Kernel</p><p>Apple’s M-series chips include hardware virtualization support and expose it through Hypervisor.framework the same technology that powers Docker Desktop, UTM, and Parallels on Mac. This framework lets a userspace process create virtual machines with hardware-backed memory isolation and near-native execution speed.</p><p>What We Had to Build</p><p>Booting an iOS kernel is not a matter of loading a binary and jumping to its entry point. XNU on iOS is deeply coupled to the hardware it was compiled for. Making it run inside a hypervisor requires reconstructing the environment it expects at every layer:</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*T_jqgUeyzh5URnY4" /></figure><p>our VM iOS kernel booting</p><p>Memory Architecture</p><p>The XNU kernel uses a split virtual address space with 16KB pages… unusual for ARM64, which typically defaults to 4KB. Our VMM constructs three-level page tables (L1→L2→L3) for both TTBR0 (identity-mapped physmap) and TTBR1 (kernel VA space), handles demand paging for the kernel’s physical memory view, and manages a descending page allocator for dynamic table allocation during boot.</p><p>What This Enables</p><p>A virtualized iOS kernel with full hypervisor-level introspection opens capabilities that have no equivalent in the physical-device world:</p><p>Enterprise Testing and CI/CD</p><p>For organizations shipping iOS apps, virtualization transforms the testing pipeline. Instead of maintaining a physical device lab with limited version coverage, you provision virtual devices programmatically. Run your test suite against iOS 15.4, 16.2, and 17.1 simultaneously on a single Mac Studio. Tear down and rebuild the environment between runs with zero hardware logistics.</p><p>Mobile Device Management Development</p><p>MDM vendors need to test enrollment flows, profile installation, and policy enforcement across dozens of iOS versions and configurations. Virtual devices make this matrix tractable: create a snapshot after enrollment, test a policy change, restore, test the next one. No manual device resets, no waiting for OTA updates, no physical USB connections.</p><p>Training and Education</p><p>Teaching iOS internals or mobile security has always suffered from the hardware problem you either need a lab full of devices or you’re limited to slides and screenshots. Virtual iOS instances let every student have their own kernel to inspect, their own debugger session, their own controlled environment for exploring how iOS actually works under the hood. Future offerings will include training on iOS specific VR workflows.</p><p>Performance and Fidelity</p><p>Because we use hardware virtualization rather than software emulation, the guest kernel executes native ARM64 instructions at native speed. There is no instruction translation, no binary rewriting, no JIT compilation. When the iOS kernel executes a load instruction, it runs on the same Apple Silicon core that powers a physical iPhone, the only difference is that page table walks go through a second level of hardware address translation managed by the hypervisor.</p><p>Where fidelity necessarily differs is at the peripheral layer. We don’t emulate the GPU, the Secure Enclave coprocessor, or the Neural Engine. Applications that depend on these subsystems won’t run, but kernel-level and system-level research, which is where the hardest security problems live, works without them.</p><p>The Technical Frontier (future state)</p><p>our iOS virtualization on Apple Silicon is still a young capability, and the frontier is moving fast.</p><p>Active areas of development include:</p><ul><li>Storage and filesystem. Emulating Apple’s ANS (Apple NAND Storage) controller to present a virtual block device, enabling full userspace boot with a mounted root filesystem.</li><li>Multi-core support. Booting secondary vCPUs to match the multi-core behavior of physical devices, important for concurrency testing and race condition research.</li></ul><p>Each of these extends the surface area of what’s possible inside the virtual machine, moving from kernel research toward full system emulation.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e8eab0d95798" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Top 10 High Impact Free and Open Source offensive tools]]></title>
            <link>https://piffd0s.medium.com/top-10-high-impact-free-and-open-source-offensive-tools-8ce1da4e4a21?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/8ce1da4e4a21</guid>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[hacking-tools]]></category>
            <category><![CDATA[pentesting]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Fri, 18 Apr 2025 19:43:35 GMT</pubDate>
            <atom:updated>2025-04-19T17:32:56.095Z</atom:updated>
            <content:encoded><![CDATA[<h4>Whether you’re a seasoned red teamer or just diving into the world of offensive security, the right tools make all the difference. In this post, I will be highlighting 10 of the most impactful, easy-to-use, and <strong>completely free</strong> tools for testing the resilience of your environment. These are the same tools used in real-world penetration tests to uncover serious gaps before real attackers do. Bold notes indicate areas where something could potentially break!</h4><p>All of these tools can be found or easily installed on any open source offensive distro like <a href="https://www.kali.org/">Kali linux</a> or <a href="https://parrotsec.org/docs/">Parrot OS</a></p><h3>1. Nmap — The Recon Workhorse</h3><p>Nmap is the backbone of any solid assessment. This network scanner helps identify live hosts, open ports, and service versions across internal or external networks. It also includes a powerful scripting engine (NSE) that can detect known vulnerabilities.</p><p><strong>Try this:</strong></p><pre>nmap -sC -sV -oA internal_scan 10.0.0.0/24</pre><p>Run default scripts, version detection, and save results in multiple formats — perfect for recon at scale.</p><h3>2. BloodHound — Map Your Path to Domain Admin</h3><p>BloodHound uses graph theory to reveal hidden privilege escalation paths in Active Directory. It’s essential for visualizing attack paths based on group memberships, session data, ACLs, and more.</p><p><strong>Collector example:</strong></p><pre>SharpHound.exe -c All</pre><p>Import the results into the BloodHound GUI to identify routes to high-value targets like Domain Admins.</p><p>The Bloodhound Community Edition is free and the GUI is containerized.<br><a href="https://bloodhound.specterops.io/get-started/quickstart/community-edition-quickstart">https://bloodhound.specterops.io/get-started/quickstart/community-edition-quickstart</a></p><h3>3. CrackMapExec — The Swiss Army Knife for Windows Networks</h3><p>CrackMapExec (CME) is a post-exploitation favorite that lets you spray credentials, enumerate SMB shares, dump hashes, and even execute commands remotely. It’s built for speed and utility inside Windows domains. <strong>Be sure to limit your password guesses per account to something that is under your account lockout policy threshold!</strong></p><p><strong>Try a credential spray with a common password:</strong></p><pre>cme smb 10.0.0.0/24 -u users.txt -p Password123!</pre><h3>4. Impacket — The Backbone of Lateral Movement</h3><p>Impacket is a Python toolkit that provides scripts to abuse common network protocols. Whether you’re dumping NTLM hashes, abusing Kerberos tickets, or remotely executing commands, Impacket is your go-to.</p><p><strong>Dump hashes using:</strong></p><pre>secretsdump.py domain/user:password@10.0.0.5</pre><h3>5. Evil-WinRM — Seamless Shell Access on Windows</h3><p>If WinRM is enabled and you’ve got credentials, Evil-WinRM gives you a powerful shell to the target system. It supports file upload/download, Powershell execution, and is ideal for post-exploitation.</p><p><strong>Connect using:</strong></p><pre>evil-winrm -i 10.0.0.5 -u administrator -p &#39;P@ssw0rd!&#39;</pre><h3>6. Responder — Easy Credential Capture</h3><p>Responder listens for LLMNR and NetBIOS broadcasts on the network and tricks systems into sending their NTLM hashes. It’s highly effective on flat networks where name resolution is misconfigured.</p><p><strong>Launch it with:</strong></p><pre>sudo responder -I &lt;ethernet interface&gt;</pre><h3>7. Burp Suite Community — Web App Hacking Made Easy</h3><p>Burp is a staple for web application testing. The Community Edition is perfect for intercepting HTTP traffic, replaying requests, and manually testing web app logic — even without automation.</p><p><strong>Setup Tip:</strong><br> Install Burp, set your browser’s proxy to 127.0.0.1:8080, and start intercepting traffic for hands-on web testing.</p><h3>8. LinPEAS / WinPEAS — Fast Privilege Esc Discovery</h3><p>These scripts automate the search for local privilege escalation opportunities on Linux and Windows. They check permissions, kernel versions, SUID binaries, stored creds, and more.</p><p><strong>On Linux:</strong></p><pre>wget https://github.com/carlospolop/PEASS-ng/releases/latest/download/linpeas.sh<br>chmod +x linpeas.sh &amp;&amp; ./linpeas.sh</pre><h3>9. Ffuf — Blazing-Fast Web Fuzzer</h3><p>Ffuf is a fast, lightweight tool for discovering hidden web directories, files, or parameters. It’s great for brute-forcing admin panels or probing API endpoints. <strong>I would reccomend testing this in QA / Dev environments first!</strong></p><p><strong>Discover hidden paths:</strong></p><pre>ffuf -u https://target.com/FUZZ -w /usr/share/wordlists/dirb/common.txt</pre><h3>10. Sliver — A Modern Open-Source C2 Framework</h3><p>Sliver is a powerful command-and-control framework built in Go. It supports multiple operating systems, encrypted comms, and even team collaboration — a true open-source alternative to commercial C2s like Cobalt Strike. <strong>Hopefully the C2 Agent gets caught by your EDR or AV!</strong></p><p><strong>Generate an implant:</strong></p><pre>sliver-server<br># then inside:<br>generate --os windows --format exe --listener http</pre><h3>Bonus!!! 11: Certipy — Weaponizing AD Certificate Services</h3><p>Certipy is an incredibly powerful tool for identifying and exploiting misconfigurations in Active Directory Certificate Services (AD CS). AD CS is often overlooked during assessments, but when misconfigured, it can offer low-privileged users a fast track to domain admin — all through abused certificate templates and enrollment policies.</p><p>Certipy supports enumeration, request forgery, authentication relaying, and even domain persistence via certificate-based authentication. If AD CS is in play, this tool is a must-have in your arsenal.</p><p><strong>Enumerate vulnerable templates:</strong></p><pre>certipy find -u &#39;user@domain.local&#39; -p &#39;Password123!&#39; -dc-ip 10.0.0.5</pre><p>Request a certificate and impersonate a domain user:</p><pre>certipy req -u &#39;user@domain.local&#39; -p &#39;Password123!&#39; -ca &#39;corp-CA\ca&#39; -template &#39;VulnerableTemplate&#39;</pre><p>Use the certificate for authentication (pass-the-cert):</p><pre>certipy auth -pfx user.pfx</pre><p>The 11th tip is one that we abuse often during internal assessments and you might be surprised how often AD CS is not audited beforehand.</p><p>Hopefully these tools will prove useful for your internal security efforts. If a more thorough assessment feel free to reach out to us at help at adversaryacademy.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8ce1da4e4a21" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Pwn2Own Miami: Aveva Edge Arbitrary DLL Loading Vulnerability]]></title>
            <link>https://piffd0s.medium.com/pwn2own-miami-aveva-edge-arbitrary-dll-loading-vulnerability-b2d10fc7d55c?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/b2d10fc7d55c</guid>
            <category><![CDATA[exploit-development]]></category>
            <category><![CDATA[penetration-testing]]></category>
            <category><![CDATA[hacking]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Thu, 01 Aug 2024 01:51:04 GMT</pubDate>
            <atom:updated>2024-08-01T01:51:04.945Z</atom:updated>
            <content:encoded><![CDATA[<p>This article covers the vulnerability I discovered while researching the Aveva Edge attack surface for Pwn2Own Miami a few years ago. It’s been patched for a while now so its fair to release my writeup of the program.</p><p>Upon testing Aveva Edge V20 SP1 and later SP2 it was found that Aveva Edge Application files (.APP) can be configured to trigger DLL loading from unsafe paths. The root cause of this condition is due to the fact that upon startup Aveva Edge will always load imgman32.dll from a safe and trusted location.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/841/1*JvULATaIuU6z0b52zVI5kA.png" /><figcaption>Imgman32.dll loaded from program files</figcaption></figure><p>However, during additional testing, it was found that imgman32.dll is responsible for attempting to load another file IMHOST32.dll however IMHOST32.dll is not included in the application package. Additionally, the call responsible for loading IMHOST32.dll is LoadLibraryA which has known abuses.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*trW9lKPmRYaGBBlJg9eCig.png" /><figcaption>Aveva Studio Manager attempting to load IMHOST32.dll</figcaption></figure><p>Analysis of the calling DLL (imgman32.dll) found the following exported function “ImgPlugInHostLoad” attempts to load IMHOST32.dll via LoadLibraryA.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/964/1*UzupkA56OTRx2Gm0q25JdA.png" /><figcaption>Exported function that calls IMHOST32.dll</figcaption></figure><p>Further analysis indicated that upon dll load imgman32.dll will also attempt to load IMHOST32.dll from dllmain. This explains why our payload (calc.exe) will spawn multiple times upon exploitation. This is a result of multiple calls to IMHOST32.dll which is backdoored in our attack payload.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*vPVPsmWrFxv_t1GSs6qs3A.png" /><figcaption>Loadlibrary of IMHOST32.dll in dllmain of imgman32.dll</figcaption></figure><p>In order to achieve successful exploitation then an attacker would need a maliciously crafted Aveva Archive, and a malicious IMHOST32.dll file alongside it. The default configuration contains code to cause imgman32.dll to load imhost32.dll</p><p>Further analysis revealed that the library responsible for loading imgman32.dll is appbuild.dll. This library leverages a number of functions from imgman32.dll all appearing to be related to handling images and image display within Aveva Edge.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/975/1*BTJBjz9TrdpyJWr0M7rnlQ.png" /><figcaption>appbuild.dll imports list from imgman32.dll</figcaption></figure><p>Further reverse engineering revealed that upon application launch the application reads TreeView.ini from a safe path. The ini file has a number of hardcoded dlls to load. This includes AppBuild.dll as the first dll that treeview.ini requires.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/623/1*Tw3wTKR4BaxPGyrV06TK0g.png" /><figcaption>Dll load list in TreeView.ini</figcaption></figure><p><strong>In closing the root cause of the issue is:</strong></p><p>· Studio Manager.exe reads TreeView.ini which contains a list of DLLs to load.</p><p>· The first loaded library in Treeview.ini (Appbuild.dll) loads imgman32.dll due to its use of imported imgman32.dll functions.</p><p>· Imgman32.dll in turn requires IMHOST32.dll which is missing from the installation path.</p><p>· Due to the fact that the Imgman32.dll uses LoadlibraryA the CWD is checked to see if IMHOST32.dll exists</p><p>· In our case an attacker has included a malicious IMHOST32.dll file with an Aveva edge “.APP” file leading to arbitrary code execution.</p><p><strong>Recommended solutions:</strong></p><p>· Include IMHOST32.dll in a trusted / preferred path</p><p>· Use a patched version of Imgman32.dll that uses a hardcoded dll path. i.e.</p><p>· HMODULE handle = LoadLibrary(“c:\\windows\\system32\\IMHOST32.dll”);</p><p><strong>Additional Findings:</strong></p><p>There were additional RCE cases in Aveva edge that were not 0-click and were therefore outside the scope of pwn2own targets.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=b2d10fc7d55c" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Geography Based Password Lists for enhanced password cracking success.]]></title>
            <link>https://piffd0s.medium.com/geography-based-password-lists-for-enhanced-password-cracking-success-91d51cd1c605?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/91d51cd1c605</guid>
            <category><![CDATA[password-security]]></category>
            <category><![CDATA[penetration-testing]]></category>
            <category><![CDATA[hacking]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Thu, 11 Jul 2024 17:59:28 GMT</pubDate>
            <atom:updated>2024-07-11T17:59:28.958Z</atom:updated>
            <content:encoded><![CDATA[<p>Recently adversary academy was on an internal penetration testing engagement. Internally we typically run things like <a href="https://github.com/lgandx/Responder">responder</a> to see what kind of hashes are available to collect. Then we pass those hashes off to our hash cracking systems for recovery to plain text. In this particular engagement there are about 200 passwords hashes collected. Using common word lists like “Rockyou” and the “realunique” list in addition to effective<a href="https://notsosecure.com/one-rule-to-rule-them-all"> Hashcat rulesets</a> some initial access was gained. However the accounts that were cracked were somewhat limited in terms of their access and privilege levels.</p><p>After having limited success with common wordlists and rulesets I turned to developing custom rulesets based on things people like in the geographic area that our client was based.</p><p>I compiled a list of popular cities, towns, activities, hobbies, sports teams and entertainment found in the geographic area in which our client was based. With a relatively small word list I saw a 33% increase in the number of cracked accounts. Those additional cracked accounts offered further access into the client environment.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/480/1*TcBxlR2FyXyst8wGortrvw.gif" /><figcaption>Thanks for the credentials!</figcaption></figure><p>With the additional access and success of the custom dictionaries I will be incorporating these types of geographic specific dictionaries for all future on site engagements for the following reasons:</p><ul><li><strong>Personal Relevance</strong>: People often use words related to their personal lives, including their hometown, places they have visited, or local landmarks. Geographic-specific words can increase the likelihood of guessing these personal passwords.</li><li><strong>Local Knowledge</strong>: If the attacker knows the victim’s location, using local words can significantly narrow down the pool of potential passwords, improving the efficiency of the attack.</li><li><strong>Predictable Patterns</strong>: People tend to create passwords using patterns that are easy to remember. This might include the names of local streets, cities, or landmarks. An attacker familiar with the local geography can exploit these predictable patterns.</li><li><strong>Publicly Available Information</strong>: Geographic information about an individual can often be gleaned from social media, public records, or online profiles. This data can be used to inform and refine password cracking strategies.</li></ul><p>If any internal or external teams attempt password cracking for either <a href="https://www.crowdstrike.com/cybersecurity-101/kerberoasting/">Kerberoastable</a> accounts or credentials that have been collected through responder I&#39;d suggest creating custom dictionaries in order to gain further access to your target environment..</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=91d51cd1c605" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[This Cloud is on Fire! Microsoft Azure Site Recovery EoP]]></title>
            <link>https://piffd0s.medium.com/this-cloud-is-on-fire-microsoft-azure-site-recovery-eop-2a8acfc3d7ee?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2a8acfc3d7ee</guid>
            <category><![CDATA[vulnerability]]></category>
            <category><![CDATA[pentesting]]></category>
            <category><![CDATA[hacking]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Mon, 13 May 2024 16:00:28 GMT</pubDate>
            <atom:updated>2024-05-13T16:00:28.647Z</atom:updated>
            <content:encoded><![CDATA[<h3>Discovering and Exploiting CVE-2024–21364</h3><p>Adversary Academy recently completed a long-term red team (assume breach) assessment for a large restaurant franchise. While performing the assessment an <a href="https://azure.microsoft.com/en-us/products/site-recovery/?ef_id=_k_Cj0KCQjw6PGxBhCVARIsAIumnWb8ObVhvcQgfKnBkvgLXGvCJvrBRPcMUvpQZ4mRNNBKpzMN35--dY4aAtFAEALw_wcB_k_&amp;OCID=AIDcmm5edswduu_SEM__k_Cj0KCQjw6PGxBhCVARIsAIumnWb8ObVhvcQgfKnBkvgLXGvCJvrBRPcMUvpQZ4mRNNBKpzMN35--dY4aAtFAEALw_wcB_k_&amp;gad_source=1&amp;gclid=Cj0KCQjw6PGxBhCVARIsAIumnWb8ObVhvcQgfKnBkvgLXGvCJvrBRPcMUvpQZ4mRNNBKpzMN35--dY4aAtFAEALw_wcB">Azure Site Recovery</a> server was found to be an attractive target in the environment. Part of our service offering is our targeted vulnerability research (TVR) program. The challenge I’ve seen with most pentest or redteam providers is that there is typically a lack of vulnerability and exploit research capabilities. Meaning if there are not known vulnerabilities with public exploit code affecting a network or environment the pentest provider can&#39;t find exploitable systems. Pentest providers typically lack full-time exploit and vulnerability development capabilities. In order to address that issue we run a program that allows our researchers to spend time attacking interesting systems they’ve encountered on customer networks, long after the engagement is over… or in this case during an engagement.</p><p>Typically on a penetration test a tester&#39;s “spidey senses” will go off at some point when you encounter a system that just feels vulnerable, or impactful if it were to be vulnerable. Our spidey senses went off when we gained access to an Azure Site Recovery (ASR) server because there appeared to be a large number of services communicating both inbound and outbound to the server as well as traffic to the customer&#39;s Azure environment. Documentation revealed that when fully deployed ASR has rights to read and write virtual machines from on-site VMware or Hyper-v systems and upload them to the Azure cloud for cloud-to-cloud disaster recovery.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/780/0*9UKlNLlS5Jed3q0M.png" /><figcaption>Our customers ASR configuration</figcaption></figure><p>While performing the engagement the research phase began immediately and we discovered a number of interesting bugs on the SRM server after we gained access to it.</p><p>Beginning our research we found that Azure SRM is site disaster recovery for one Azure region to another region or physical to the cloud.</p><p>SRM can replicate on-premises hypervisors VMware, Hyper-V, physical servers (Windows and Linux), or Azure Stacks to Azure sites.</p><p>Basically, Microsoft said, “We will support anything other than AWS or GCP!”</p><p>As we started our research we found roughly 20 previous CVEs affecting Microsoft Azure SRM, most were EoP and most were found in 2022. Hopefully, we could find something new.</p><p>Our research mindset typically includes mapping out application behaviors and what could go wrong with misconfigurations, logic flaws, or historically problematic issues (in this case EoP).</p><p>We started by reviewing features, capabilities, and processes in Azure SRM and found that:</p><ul><li>the SRM process and config server runs a web server listening for replication events to the backup server on port 9443</li><li>Process server must have permission to read all properties from all systems being backed up</li><li>Process server must have the ability to read/write to Azure for synchronization, and deployment of agent</li><li>SRM server connects to clients via WMI/ credentials stored in the DB</li><li>This WMI connection deploys the SRM mobility agent responsible for the agent to server comms.</li></ul><p>Once this behavior was documented we decided that the web server privileges might be important, and the WMI credentials stored in the local database were definitely valuable targets to begin attacking.</p><p>Reviewing files accessed on startup by the services showed us that a config file named amethyst is read on startup. Here was the first bug we found.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*tyiIJRUKBU2NaY_N" /><figcaption>The configuration file is readable by any local user account on the SRM server</figcaption></figure><p>The amethyst config file contains the plaintext mysql DB root username and password, this allows us to interact with the local database as root.</p><p>Connecting to the mysql database we began to debug and monitor the mysql queries that were executed by the server. Here we found our attack target.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*PLZ5J6rb0WiY7ZSy" /><figcaption>The Azure SRM credentials are stored AES encrypted in the database. The encryption key is not readable by a user.</figcaption></figure><p>We found php code responsible for executing the query that decrypts and uses the credentials we want access to. The first roadblock encountered is that we are not able to read the Encryption.key file as a standard user.</p><p>After some research and failed attempts, we found a Solution!</p><p>If the process responsible for handling the php / mysql queries has access to the key, we must become the process.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/0*IVV9yW6r3cPEl-RO" /><figcaption>php-cgi reading the encryption.key</figcaption></figure><p>As our standard user account on the server, we don’t have the SeImpersonatePrivilege, we don&#39;t have an admin account on the server either. So we needed to find a bug affecting the web server.</p><p>Further research allowed us to find a directory on the server where the web server / php code isn’t properly secured. We can write a webshell to this directory and “become” the web server process.</p><ul><li>The web services are running as IUSR which DOES have the SeImpersonatePrivilege</li></ul><figure><img alt="" src="https://cdn-images-1.medium.com/max/672/0*-4yICHGTcV2SOjJR" /><figcaption>Spawning a beacon as the IUSR user from our web shell</figcaption></figure><p>We then can use SEImpersonatePrivilege to read the encryption.key</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/794/1*-Xd3D9DEg0iJwXZDlK3XHw.png" /><figcaption>Impersonating the user we want to access the encryption.key</figcaption></figure><p>The final challenge was overcoming some weird character-handling behavior by MySQL which can&#39;t handle the characters in the encryption.key inline, so store it as a variable to use the key and decrypt the admin credentials.</p><p>After discovering the bugs and disclosing the credentials used by SRM the team was able to access the Vsphere environment, took snapshots of the domain controllers, and performed offline attacks to recover Enterprise and Domain admin access. After exploiting the issues we reported the vulnerability to Microsoft and received recognition for <a href="https://msrc.microsoft.com/update-guide/vulnerability/CVE-2024-21364">CVE-2024–21364</a> with a patch becoming available several months later.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2a8acfc3d7ee" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Clocking into The Network — Attacking KronosTimeclock Devices]]></title>
            <link>https://piffd0s.medium.com/clocking-into-the-network-attacking-kronostimeclock-devices-d7b226a692f3?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/d7b226a692f3</guid>
            <category><![CDATA[pentesting]]></category>
            <category><![CDATA[hacking]]></category>
            <category><![CDATA[red-team]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Wed, 27 Sep 2023 17:15:28 GMT</pubDate>
            <atom:updated>2023-09-27T17:15:28.037Z</atom:updated>
            <content:encoded><![CDATA[<h3>Clocking into The Network — Attacking KronosTimeclock Devices</h3><p>One of the services I’m most excited about at <a href="https://adversaryacademy.com/">Adversary Academy</a> is our targeted vulnerability research (TVR) program. The challenge I’ve seen with historical pentest providers is that there is typically a one to two-week engagement window and after that, you usually don’t hear from your pentest provider until it&#39;s time for next year&#39;s test. In order to disrupt that cycle We’ve started a program that allows for researchers to spend time attacking interesting systems they’ve encountered on customer networks, long after the engagement is over. Typically on a penetration test your “spidey senses” will go off at some point when you encounter a system that just feels vulnerable, or impactful if it were to be vulnerable. With the TVR program, we are able to spend cycles researching those items that appear to be high-impact.</p><p>One recent example was for a customer who employed the Kronos InTouch DX timeclock device. A really fancy Android based timeclock that supports biometric data as well as facial recognition.</p><p>For this engagement, the ability to jump to an enterprise network would be very valuable. We hypothesized that the Kronos Timeclock devices may be connected to an enterprise network and not properly segmented. Attempting to access all of the settings on the customer devices were locked out and non-default passwords were used. Later we purchased our own version of the hardware to perform a full teardown.</p><p>Further documentation available on the <a href="https://fcc.report/FCC-ID/P5W-8610K007/4454950.pdf">FCC report website</a> shows that ssh is an option that can be configured on the device and that a maintenance mode badge or button can be used to bypass the initial configuration.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/612/1*JZpo0smqLwAl9fjFFrkznw.png" /><figcaption>Pressing and holding the maintenance mode button (4) will bypass the locked screens</figcaption></figure><p>After bypassing the lockout and enabling SSH with a known password the user is logged in as root rather than a low privilege account. With root access, any configuration can be changed however one of the most valuable configuration files we found was the wpa_supplicant.conf file which contains the wifi credentials in plaintext. This file would allow a local attacker to then join the network that the Kronos time clock is connected to, potentially joining an enterprise network and carrying out further attacks once on the network.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/851/1*8FJNDQIm0AwSb-13h3bavQ.png" /><figcaption>our root access as rauser printing out the wpa_supplicant.conf file</figcaption></figure><p>After discovering this issue the Adversary Academy team reached out to Kronos and recommended that the wpa_supplicant file be wiped when maintenance mode is entered by physical keypress, which would better protect the wpa_supplicant.conf file, or alternatively making the Rauser non-root and only giving them access to the needed configuration files is also a suitable alternative. Currently no patch or update is available for this issue and we have yet to hear a response from Kronos / UGK.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=d7b226a692f3" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[No Mud No Lotus: The art of transforming breaches through security improvement]]></title>
            <link>https://piffd0s.medium.com/no-mud-no-lotus-the-art-of-transforming-breaches-through-security-improvement-2b21f0ba7161?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/2b21f0ba7161</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[security]]></category>
            <category><![CDATA[hacking]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Thu, 09 Mar 2023 21:52:59 GMT</pubDate>
            <atom:updated>2023-03-09T21:52:59.474Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*brNxDxXJUuDsO8Ddu-1F3A.jpeg" /><figcaption>A lot of gross stuff at the bottom of a pond is responsible for this</figcaption></figure><p>In Buddhist philosophy I often hear the expression “No Mud, No Lotus” this expression aligns with the Buddhist view that life and existence in many ways are circular. Things that are negative can actually be used for our benefit, and things that are good when overused can harm us. Life is a duality.</p><p>I&#39;ve been thinking about how organizations are transformed by negative events, specifically security breaches or incidents. These unfortunate events which are caused by evildoers with malicious intent usually have some unintended consequences. Those unintended consequences are, hopefully, that the affected organization&#39;s cybersecurity posture will significantly improve as a result of the breach. Unfortunately, sometimes the evildoer&#39;s intended purpose, financial loss, or monetary gain is also the outcome.</p><p>In the case of improvements in cybersecurity after a breach, this improvement is not without much pain and suffering on the part of the incident responders who work countless hours, the customers of the organization who lose their data, or their identities, and the IT staff who have to rebuild after much is destroyed.</p><p>Often times out of the ashes of a significant breach, a more defensible organization with a more realistic view of the cost of failure is born. CISOs, executives, and board members who have never before been leaders in an organization that has been hit by a massive incident now understand that security is truly everyone&#39;s responsibility. These lessons, in the case of a breach, are learned the hard way. “The hard way” is certainly a way but it&#39;s not always the best way, and it&#39;s a way I would like people to avoid if at all possible.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*gjxx1ozNvIL_R3VEozAXlg.gif" /><figcaption>Your SOC if they haven’t dealt with an adversary before</figcaption></figure><p>Returning back to Buddhist philosophy there is also the concept of the “middle way.” Applying that concept to cybersecurity informs us that it may be possible to significantly improve security without the pain of a significant breach. What would that look like? Well, from my perspective the middle way does still have an adversary, just not one that wants to cause you actual monetary or other harm. Many cybersecurity thinkers have quoted the art of war by Sun Tzu, a 6th-century BCE military strategist. I will spare you recitations of those concepts. Rather, in keeping with Buddhist inspiration I will explore a few concepts from “<a href="https://www.goodreads.com/book/show/867247.A_Book_of_Five_Rings">The Book of Five Rings</a>” by Miyamoto Musashi. Miyamoto was an <strong><em>incredibly</em></strong><em> </em>skilled Japanese swordsman, philosopher, strategist, writer, and rōnin. (I bet you thought your LinkedIn profile was impressive!) Miyamoto was also a Buddhist, not necessarily in the peace-loving modern sense, this was feudal Japan and Miyamoto killed <strong>a lot </strong>of people… but I digress… how can his strategies and philosophies help defend the modern enterprise?</p><p>Taking inspiration from The Book of Five Rings here are a few quotes to ponder.</p><h3>“The important thing in strategy is to suppress the enemy’s useful actions but allow his useless actions”</h3><figure><img alt="" src="https://cdn-images-1.medium.com/max/640/1*H1rNLlDnWtI7yf2WqtHbcw.gif" /><figcaption>Miyamoto dropping knowledge bombs</figcaption></figure><p>Here are the factors to consider:</p><p>When an adversary gains a foothold in an environment <em>we need to ensure that they are not able to take any useful actions without being detected and blocked</em>. There are however a number of useless actions that we can allow them to take on the system, this allows them to waste their time on a system, and increase the likelihood that they will be detected as soon as they attempt a useful action. Using a tool to automatically evaluate your EDR frameworks detective capability comes to mind. A tool like MITRE’s <a href="https://caldera.mitre.org/">caldera framework</a> can launch a battery of tests on an endpoint. You can then evaluate which actions your EDR solution can detect, which it cannot detect, which undetected actions should be prioritized, and which ones can safely be ignored. If you do this <strong>you will be implementing Miyamoto&#39;s strategy of suppressing the enemy&#39;s useful actions.</strong></p><h3>“You can only fight the way you practice”</h3><p>After running countless purple team engagements, red team exercises, and penetration tests over my career<strong> <em>there has not been a single time that all teams did not collectively walk away with something to improve on or focus on for the next time</em>.</strong> If you are not practicing how to defend your environment from an adversary on a regular basis how can you expect to fight one off in a real-world breach?</p><p>There are so many other insightful quotes in the Book of Five Rings but I will leave you with one final gem.</p><p><strong>“In strategy, it is important to see distant things as if they were close and to take a distanced view of close things.”</strong></p><p>This one should hit home for all security people, there is a constant flood of small tasks, alerts, things to do, and people to help. Those are close things, consider the urgency of someone reporting a phishing email. Yes, it is possibly a phishing email, but what if you were to wait to respond for 30 minutes and plan out an incident tabletop? The distant things (a breach) need to be examined closely, what are you doing to prepare for that eventuality?</p><p>okay okay, last one:</p><h3><strong>“You must understand that there is more than one path to the top of the mountain”</strong></h3><p>There is no one right way to secure an organization, and there are also many wrong ways. Many organizations choose a variety of paths to improved security, some build out their own adversary emulation teams, and others bring in an outside party, some keep their systems disconnected from the internet entirely. If you’d like to discuss what may work for your organization you can reach me at chris [at] adversaryacademy.com</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=2b21f0ba7161" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Confessions of a bug bounty program manager]]></title>
            <link>https://piffd0s.medium.com/confessions-of-a-bug-bounty-program-manager-3fc975918456?source=rss-796ac520f43c------2</link>
            <guid isPermaLink="false">https://medium.com/p/3fc975918456</guid>
            <category><![CDATA[bug-bounty]]></category>
            <category><![CDATA[information-security]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Chris Hernandez]]></dc:creator>
            <pubDate>Fri, 17 Feb 2023 23:47:08 GMT</pubDate>
            <atom:updated>2023-02-17T23:47:08.176Z</atom:updated>
            <content:encoded><![CDATA[<p><a href="https://piffd0s.medium.com/confessions-of-a-top-ranked-bug-bounty-hunter-5f4d71d34598">In my previous article</a> I wrote about my experiences as a top ranked bug bounty hunter. In this article I will write about my experiences on the other side of the fence triaging bug bounty program submissions. This article will hopefully serve to highlight some of the traps that exist in the bug bounty space.</p><p>Hopefully my dual roles in the bug bounty space will help existing researchers have insight into what a program manager might be looking for when receiving a bug bounty submission. I am also hoping to I highlight some of the challenges that I see for the customer or client side of the bug bounty space.</p><p>When I was supporting a bug bounty program the program was managed by B<a href="https://bugcrowd.com/">ugcrowd</a>. Other programs may have their own unique set of challenges, but I would expect that some of the common challenges are exactly the same across the majority of the bug bounty space.</p><p>Looking back from a value perspective, I think that the highest value we received was from two unique events happening in our program.</p><p>T<strong><em>he first event </em></strong>was<strong><em> </em></strong>during the initial program launch. New programs usually attract quite a bit of attention as researchers rush to find shallow and easily monetized bugs. As a result of this attention the initial “penetration test” or targeted test phase of the bug bounty program revealed a decent number of low to medium severity findings. If a company were looking to replace a traditional point in time penetration test and still be able to provide auditors with a “pentest report” this initial launch phase would be a good place to look to replace a traditional pentest vendor albeit at a higher price point.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/492/1*HzeEdVsGXO3D5NuTdz7Pjg.png" /><figcaption>Quick! find XSS!</figcaption></figure><p><strong>However</strong>, once the initial burst of activity was over quickly delved into an extended period of low value or duplicate bugs. Something <strong><em>a company might want to consider is the amount of time that it takes their internal staff to triage and review bugs that are essentially useless</em></strong><em>.</em> During this time, I found myself to be frustrated with the program. I can’t blame the researchers however, as I mentioned in my previous article the economics of most bug bounty programs incentivizes researchers to target easily automated bugs.</p><p>T<strong><em>he Second Event </em></strong>was quite a bit later in the programs run time when we finally had researcher submit high impact bugs. These bugs demonstrated the value of having access to a large researcher pool. As I mentioned before, instead of focusing on shallow bugs and easily automated bugs some researchers focus on a specific class of vulnerability. And in our case, we were able to attract the attention of a researcher who had familiarity with our tech stack.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/600/1*D6NUzVUgniE1ifuVUXytxw.jpeg" /><figcaption>Bug Bounty! Great Success!</figcaption></figure><p><strong>However,</strong> one thing to consider is the average price of a yearly contract with a bug bounty vendor is about $100,000. So, while the high impact vulnerabilities were valuable I still have difficulty matching the value extracted to the price paid per year. As usual your mileage may vary with a program like this, I would expect to see the number of high and critical vulnerabilities trail off year over year. So it may be that the first year or two of the bug bounty programs represents a great value for your organization and in the years following the value starts to trailer in correlation with the number of high and critical submissions.</p><p><strong>So, what can someone do to ensure their program is successful?</strong></p><p><em>If it were up to me, these are the questions that I would ask a bug bounty vendor.</em></p><p><strong>Do you have a certain number of researchers who are skilled in my tech stack?</strong></p><p>Bounty vendor may have statistics on the number of researchers who are part of their program who are skilled at say API testing. But do they have details on the percentage of researchers who are skilled at finding Java deserialization vulnerabilities? Those types of metrics might be useful for someone who knows what they’re using for their tech stack.</p><p><em>If they can’t answer this question with explicit detail, you may want to consider other options!</em></p><p><strong>Can I invite or incentivize certain researchers?</strong></p><p>If someone is finding bugs in your platform, you may want to ask them and incentivize them to spend more time looking at your software.</p><p><strong>Realize that most bug bounty programs do not have exclusive access to researchers. </strong>Most researchers are looking at programs across multiple platforms. So, if you can get a better price with one vendor versus another it may be in your best interest to go with that vendor.</p><p><strong>Pay above an above average rate for high and critical vulnerabilities.</strong></p><p>Really, we are trying to reward researchers for their time and disincentivize researchers for low impact bugs. If it were up to me, I would pay a below average payout for low and medium severity bugs and an above average payout for high and critical bugs. In the screenshot below you can see Bugcrowd’s payment trends by severity. For P1 severity bugs the average payout is $1000 to about $5000. I would probably start p1 pay outs at $5000 to $10+ thousand depending upon the overall impact to the organization.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*5J2FENVX-egwdP5rzXssjw.png" /><figcaption>make your p1 and p2 payouts above the average line, p4 and p5 below average</figcaption></figure><p>Think about it this way, if you pay bugcrowd $100,000 for one year of a bug bounty program, and you only get one P1 bug because your payout was at or below the average payout. That bug essentially cost you 100k, yet the researcher only gets 1 or 2k for their work. If instead, you set the payout to $10,000 and you get 10 p1 bugs, you are paying the market rate for exports and significantly driving down your overall attack surface. In my mind that is the definition of a win-win. Researchers are getting paid for their time and you are getting a better return on investment.</p><p>Are there any other options?</p><p><a href="https://adversaryacademy.com/">Adversary Academy</a> offers targeted vulnerability research as part of its pentest services. This breaks the traditional challenge with penetration tests being point-in-time assessments and brings in the perspective and capabilities of an advanced and well funded adversary. In this case you are paying for sustained and focused research targeting your enterprises technology.</p><p>Another option would be a partnership with a responsible vulnerability acquisition platform like ZDI. For some pwn2own categories you can sponsor or support the event and your software or hardware will be in the list of targets for the competition. <a href="https://www.youtube.com/watch?v=ZUs98Z-pIpY">Tesla does this every year</a>.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=3fc975918456" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>