<?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 Brent Hachey on Medium]]></title>
        <description><![CDATA[Stories by Brent Hachey on Medium]]></description>
        <link>https://medium.com/@brent.hachey?source=rss-2a05f76d47f0------2</link>
        <image>
            <url>https://cdn-images-1.medium.com/fit/c/150/150/1*RCnqnAmTlL9EAUnDwkavKQ.jpeg</url>
            <title>Stories by Brent Hachey on Medium</title>
            <link>https://medium.com/@brent.hachey?source=rss-2a05f76d47f0------2</link>
        </image>
        <generator>Medium</generator>
        <lastBuildDate>Sun, 22 Nov 2026 09:11:30 GMT</lastBuildDate>
        <atom:link href="https://medium.com/@brent.hachey/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[A Threat Hunt Report for the VMware vCenter Zero Day Exploited by ARMORY CASTLE]]></title>
            <link>https://medium.com/@brent.hachey/a-threat-hunt-report-for-the-vmware-vcenter-zero-day-exploited-by-armory-castle-930fe121e4fa?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/930fe121e4fa</guid>
            <category><![CDATA[threat-hunting]]></category>
            <category><![CDATA[ai-agent]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Fri, 20 Nov 2026 14:12:56 GMT</pubDate>
            <atom:updated>2026-11-21T00:34:12.270Z</atom:updated>
            <content:encoded><![CDATA[<figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*Bi-09CgRWnNNGQr0-RpGmQ.jpeg" /><figcaption>ARMORY CASTLE vCenter Zero Day</figcaption></figure><p>THREAT HUNT REPORT<br>VMware vCenter zero day exploited by ARMORY CASTLE (CVE-2026-123890), Nov 2026<br>Written by Brent Hachey</p><p>Facts as of Nov 20, 2026, about 08:30 UTC. The story is new, so recheck the sources before relying on it.</p><p>HOW TO READ THIS REPORT</p><p>This is a reconstruction built from public reporting. It was not run against any environment. Each section has two parts. &quot;This report&quot; records the facts for this incident and says who observed them. &quot;In your report&quot; says what to write in your own version and where the information comes from. Examples are marked as examples. Format blocks are blank on purpose and contain no real data.</p><p>---</p><p>HANDLING</p><p>This report: Public. Built from public reporting. No employer data and no internal telemetry.</p><p>In your report: apply your own classification and retention rules. Redact hostnames, usernames and internal identifiers before sharing outside your team. Treat raw logs, screenshots and credential artifacts as restricted.</p><p>---</p><p>HUNT DETAILS</p><p>This report: Prepared Nov 20, 2026. Not run, so no start or end date. Lead analyst: Brent Hachey.</p><p>Data sources needed, and what each one answers:<br>- vCenter appliance logs (directory service vmdird, SSO and VAMI). Account creation, SSH enabled through the management API, sessions as administrator@vsphere.local.<br>- vCenter event and audit history. Who changed what, and when.<br>- Remote syslog from the vCenter appliance. Google says the VAMI firewall does not send remote logs for denied connections or shell activity, so without remote syslog this is a blind spot.<br>- Network flow or firewall logs for traffic to vCenter and ESXi. Who connected to TCP 2012, 5480, 22 and 443, and any host that scanned several VMware ports.<br>- Telemetry from the Linux appliances and servers that can reach vCenter. Web shells, files and keys in /tmp, SUID changes, outbound Slack calls. Google says the appliances in this case had no EDR.<br>- BeyondTrust Remote Support appliance logs and patch state. The entry point, and where Google says staged data was likely pulled out.<br>- ESXi host logs. SSH and Host Client logins, changes to root&#39;s authorized_keys.<br>- Change tickets for SSO accounts, groups and SSH changes. The way to tell a real change from an attacker&#39;s.</p><p>Keep history back to early August where you can. Log file locations differ by vCenter version and are not given in the sources, so confirm them for your version.</p><p>In your report: dates, lead and supporting analysts, and the data sources you actually used with the retention you had for each.</p><p>---</p><p>EXECUTIVE SUMMARY</p><p>This report: On Nov 19, 2026, Broadcom patched CVE-2026-123890, a 9.8 authentication bypass in the VMware Directory Service in vCenter. Its advisory (VMSA-2026-0009) does not mention attacks. Google&#39;s Mandiant and GTIG say the suspected PRC nexus actor ARMORY CASTLE (formerly UNC5221) exploited it as a zero day in early September 2026, more than two months before the patch. In the intrusion Google describes, the group came in through a BeyondTrust Remote Support zero day, moved to a Linux appliance with no EDR, ran a custom tool against vCenter, forged SAML tokens, and ended on an ESXi host reading files out of a guest disk. This report turns Google&#39;s details into a look back at your vCenter, the Linux systems around it and your ESXi hosts, and it separates what was reported, what is inferred and what is not known. Patching closes the door but does not remove what an attacker already took.</p><p>In your report: three to five sentences on what you examined, what you found and why it matters.</p><p>---</p><p>OUTCOME STATUS</p><p>This report: Not run. Nothing was tested against an environment, so there is no result for either hypothesis and no finding to record.</p><p>In your report: give a result for each hypothesis and for each finding, and say why for each. One overall result is not enough, because a hunt can end differently in different places. Use Confirmed when your logs show a match you cannot tie to a person or a ticket. Use Refuted when you had full coverage and found no match. Use Inconclusive when gaps in the data section prevent either answer. Example format only. H1 on vCenters with full logs, Refuted. H1 on a vCenter where appliance logs rolled over, Inconclusive. H2 on Linux appliances with no telemetry, Inconclusive. No findings is still a finding, so record every result, including the negative and partial ones.</p><p>---</p><p>USE CASE ALIGNMENT</p><p>This report: Use Case: compromise of virtualization management infrastructure. Category: post compromise, identity and management plane. ATT&amp;CK: see the ATT&amp;CK Mapping section.</p><p>In your report: the use case name your program uses, its category, and the techniques the hunt tests.</p><p>---</p><p>HYPOTHESIS</p><p>This report has two hypotheses because the two tests need different data.</p><p>H1, your vCenter and ESXi: If an adversary used CVE-2026-123890 against a vCenter before it was patched, the vCenter will show an SSO administrator account, an SSH enable and disable through the management API, or an administrator session that cannot be tied to a person or a ticket, and ESXi hosts will show logins or key changes that match.</p><p>H2, the path to vCenter: If the adversary reached vCenter from inside the network, a Linux host near vCenter will show scanning for VMware ports, executables and keys in temporary directories, and web shells, and vCenter&#39;s SSO control port will show connections from a source that is not part of normal management traffic.</p><p>Core Principle: All malicious activity is anomalous, but not all anomalous activity is malicious. Administrators do create accounts, change groups and enable SSH, so a change alone is not enough. It has to fail to match a person or a ticket.</p><p>In your report: one statement per hypothesis covering the scenario, the purpose and what malicious activity would look like if present.</p><p>---</p><p>INTELLIGENCE AND CONTEXT</p><p>This report: sources, and how far to trust each.</p><p>1. Google Threat Intelligence Group and Mandiant, &quot;ARMORY CASTLE Exploits vSphere&quot;, Nov 19, 2026. Primary for the exploitation. Read in full from the page text. Observation by the vendor that investigated, with forensic evidence for the vCenter exploitation. Mandiant and GTIG also recommend their own hardening tool, so that advice is from an interested party. The report describes one intrusion. Some tooling is described as &quot;likely&quot; used, and the exfiltration to the actor is stated as &quot;likely&quot;, not seen.</p><p>2. Broadcom advisory VMSA-2026-0009, Nov 19. Primary for the vulnerability. Describes an authentication bypass in the VMware Directory Service in vCenter, Critical, CVSSv3 9.8. Lists fixed versions, lists no workaround, and does not mention exploitation in the wild. NVD (Nov 19) mirrors the vendor data and adds nothing independent.</p><p>3. SecurityWeek, Eduard Kovacs, Nov 19. Secondary, relays Google. Adds that Mandiant&#39;s CTO said the company investigated multiple incidents and that organizations should patch and examine systems for compromise. Google&#39;s report describes one.</p><p>4. BleepingComputer, Lawrence Abrams. Secondary, relays Google. It states the BeyondTrust zero day was exploited since late August. That date is not in Google&#39;s Nov 19 report, which points to an earlier Oct 15 blog, so it is held here as one outlet&#39;s claim.</p><p>5. The Hacker News, Swati Khandelwal. Secondary. Says the Google analysis was published Nov 20. Google&#39;s own page is dated Nov 19, so that detail was wrong or a time zone effect.</p><p>In your report: list each source with its date and link, say whether you observed it or read it, and note any limits or interests.</p><p>---</p><p>ENVIRONMENT SCOPE</p><p>This report:</p><p>In scope:<br>- Every vCenter Server you run, including those inside VMware Cloud Foundation, vSphere Foundation and Telco Cloud products.<br>- The Linux appliances and servers that can reach vCenter or ESXi, especially any without EDR.<br>- Any BeyondTrust Remote Support appliance.<br>- ESXi hosts managed by those vCenters, and the guest virtual machines whose disks they hold, since the actor read files straight out of a guest disk.<br>- Identity systems that trust vCenter SAML tokens.</p><p>Why this window. Google dates the vCenter exploitation to early September, the BONESAW build timestamp to Jul 13 (Go build metadata can be modified), and the actor&#39;s knowledge of the flaw to before Jul 29. BleepingComputer dates the BeyondTrust exploitation to late August. Start at Aug 1 if retention allows, and no later than Aug 20, and run to the day you patched.</p><p>Deliberately out. Guest operating systems except where a guest disk on an ESXi host could have been read. Anything you do not run.</p><p>In your report: systems, accounts and segments included, and each exclusion with its reason.</p><p>---</p><p>BASELINE DEFINITION (30 TO 90 DAYS)</p><p>This report: A baseline ending before the window opens (Jul 31 or earlier).</p><p>Identity. The SSO users and groups that existed, who created them and when. The members of the administrator groups. The normal sources of administrator logins.</p><p>Network. Which hosts normally talk to vCenter&#39;s management and SSO ports, and which Linux hosts ever talk to ESXi.</p><p>Management actions. How often SSH is enabled on vCenter and ESXi, by whom and from where.</p><p>Linux hosts. The binaries that normally run from temporary directories (usually none), and which hosts run nmap.</p><p>In your report: what normal looks like, set before you evaluate anomalies, and the dates it covers.</p><p>---</p><p>REQUIRED DATA VS. ACTUAL DATA</p><p>This report: Not run, so there is no actual data to compare. What the hunt would need before it starts:<br>- Retention back to early August for every source under Hunt Details, and change tickets to compare against.<br>- How long vCenter and directory service logs are kept, since appliance logs can roll over.<br>- Whether vCenter logs leave the appliance.<br>- Whether flow data shows traffic to TCP 2012.<br>- Whether the Linux appliances near vCenter have any telemetry at all. In Google&#39;s case they had none.</p><p>Without logs you can check current state (accounts, groups, certificates, authorized keys) but cannot show when it changed.</p><p>In your report: what telemetry the hypothesis needs, what you actually had, the gaps, and your confidence in each.</p><p>---</p><p>TOOLS AND AI USE</p><p>This report: Tools. Public web sources only. No SIEM, EDR or other platform was used because nothing was run. AI. No hunt was run, so no AI was used to run one and no telemetry was interpreted.</p><p>In your report: list every tool used to run the hunt, then state plainly whether AI factored in, which tool, which task, and that a human validated its output before it reached the findings. If none, say none.</p><p>---</p><p>METHODOLOGY (UNTESTED)</p><p>This report: logic and windows for each step. Nothing here was run. The steps are the analyst&#39;s ideas built from Google&#39;s reporting. Translate them to your platform and record the actual queries.</p><p>Step 1. Inventory your vCenters, versions and patch state against VMSA-2026-0009. Note any on 7.0, which has no public patch listed.</p><p>Step 2. List SSO users and groups. Find any account or group membership created or changed since the window opened and compare each against a person or ticket. Google reports a backdoor machine account created through vmdir, a local administrator account named sps followed by a redacted string at vsphere.local, and Global Administrator rights assigned on the vCenter root folder.</p><p>Step 3. In vCenter and VAMI logs, look for a PUT to /rest/appliance/access/ssh that enabled SSH and a disable shortly after, and for SSH logins by an account you do not recognize. Google says the actor enabled SSH, connected, took the files, then disabled it.</p><p>Step 4. In flow or firewall data, find sources that connected to TCP 2012 on vCenter that are not part of your management traffic. Look for internal scanning of 2012, 9080, 5480, 443, 389 and 22 from one host.</p><p>Step 5. On Linux appliances near vCenter and any remote support appliance, look for web shells (ActiveSer.jsp, u.jsp), files in /tmp that match the names below, the SUID bit set on /bin/env, gdb attached to a running process, base64 text echoed into a file and run, curl with -k writing to /tmp, an nmap run, and outbound traffic to Slack from a server.</p><p>Step 6. On ESXi, review Host Client and SSH logins (the actor used the vpxuser account and root), changes to root&#39;s authorized_keys, hidden directories on VMFS volumes (the actor used one named .a), and a Python tool run against disk devices.</p><p>Step 7. Check the SSH banner SSH-2.0-pulsesecure_23345 on any listening host, and listeners on TCP 53 and 33992 on Linux appliances.</p><p>Step 8. For any BeyondTrust Remote Support appliance, confirm the BT26-05 patch date for CVE-2026-102764 and review the appliance and its neighbors for the same period. Google says it likely served as the exfiltration point.</p><p>In your report: the structured process in sequence, with the actual queries and the logic and time window for each. Example from the original template: &quot;Queried EDR platform for process creation events where parent process equals winlogon.exe, 90 day window.&quot;</p><p>---</p><p>TECHNICAL EVIDENCE</p><p>This report: public artifacts, each credited to Google and Mandiant. No artifact was observed by this report&#39;s author.</p><p>The flaw. CVE-2026-123890 is an unauthenticated SRP 6a session key disclosure over DCE/RPC in the vCenter directory service (vmdird), through a cryptographic logic and authorization flaw in the rpc_srp_verifier service over port 2012, helped by predictable UUIDv1 timestamp generation. It yields an authenticated session as administrator@vsphere.local, then command execution and file access, including data.mdb through the vmdirdbcp RPC interface.</p><p>Reconnaissance. nmap -sS scans for ports 443, 2012, 389, 445, 3389, 22, 21, 5480 and 9080 on internal subnets, and visits to the URIs /sdk and /ui.</p><p>Foothold. Web shell ActiveSer.jsp (SLIMSAVAGE) with the parameter pwd set to 123. About four hours later web shell u.jsp (CAGEMATCH) for staging files into /tmp.</p><p>Privilege escalation. SUID set on /bin/env through gdb, driven by a base64 string echoed into /tmp/gdb_file.</p><p>Tooling. A RAR binary and an archive 123.rar fetched with curl from the Remote Support appliance (path content/r.js), extracted to /tmp/123 (BONESAW).</p><p>BONESAW. Go utility, named by the actor vcenter-db-teamcheck-go, with a csa subcommand to create an SSO administrator. Writes files such as anon host data.mdb and tenant credential key, leaf and chain files under a temp directory. Build timestamp 2026-07-13T09:21:24Z.</p><p>SAWSNAKE (/tmp/sss). Go port of a public vCenter SAML login tool. Extracts IdP signing keys and certificates from data.mdb, forges SAML assertions for the Administrator account and obtains administrator session cookies. Output files idp_key.pem, trusted_cert_2.pem, domain.txt. Deleted with rm after use.</p><p>BEARBONES (/tmp/dbr). C backdoor based on Dropbear SSH. Bypasses password authentication, logs plaintext credentials, loads host keys from /tmp. Banner SSH-2.0-pulsesecure_23345. Run as dbr -R -p 53 and dbr -R -p 33992.</p><p>vCenter actions. SSH enabled through the VAMI API (PUT to /rest/appliance/access/ssh), login with a new local administrator account, theft of the internal master crypto key file and the PostgreSQL password file, SSH disabled right after.</p><p>ESXi actions. Login to the Host Client as vpxuser and by SSH as root. WinSCP to browse. A public key added to root&#39;s authorized_keys. A hidden directory .a on a VMFS volume. A Python tool (vmfs_browser.py) used to extract NTDS, the SYSTEM and SECURITY registry hives and DNS files from a guest virtual machine disk. Results archived as o.tar.gz, copied out, directory deleted.</p><p>Exfiltration. Staged archives were downloaded by HTTP GET from the Remote Support appliance, from where Google says the actor likely exfiltrated them.</p><p>LAZYBONES (/usr/bin/vmtoolscs). Go backdoor using Slack for command and control with AES 256 CTR, restarted by a systemd service every 50 minutes. Seen on an additional Linux appliance with /tmp/123.key. Seen before by Mandiant in a separate ARMORY CASTLE intrusion in early 2025.</p><p>In your report: raw artifacts such as event IDs, command lines, log excerpts and ticket patterns. Format only, no real data: Source | Timestamp UTC | Host or user | Account or process | Action | Result.</p><p>---</p><p>BEHAVIORAL SUMMARY</p><p>This report: Per Google, in order.<br>1. Entry. Came in through a BeyondTrust Remote Support zero day.<br>2. Foothold. Moved to an internal Linux appliance with no EDR, dropped web shells, and escalated to root.<br>3. Reconnaissance. Scanned for VMware ports, including 2012.<br>4. vCenter. Ran BONESAW to get an administrator session, take the directory database and the SAML signing key, and create an SSO administrator account. Forged administrator tokens with SAWSNAKE.<br>5. Persistence. Kept access with web shells and a Dropbear based backdoor.<br>6. Credential theft. Enabled SSH through the management API, took the master key and the PostgreSQL password, then turned SSH off.<br>7. ESXi. Logged in, added its own SSH key, and used a Python tool to read files out of a guest disk, including NTDS and registry hive files.<br>8. Exfiltration. Archived the files and pulled them out through the Linux appliance and the remote support appliance (Google says likely).<br>9. Clean up. Deleted its tools and staging directories as it went.</p><p>In your report: what your environment did, synthesized from the evidence.</p><p>---</p><p>FINDINGS</p><p>This report: no findings, because nothing was run. The public record separates into four lists.</p><p>Reported by Google and Mandiant.<br>- Everything under Technical Evidence.<br>- Exploitation in early September 2026 against a vCenter already patched for CVE-2026-59309.<br>- A BONESAW build timestamp 16 days before CVE-2026-59309 was disclosed (Jul 29), and Mandiant&#39;s evidence that the actor knew of CVE-2026-123890 before that date. Go build metadata can be modified, and Google says so.<br>- Attribution to the suspected PRC nexus actor ARMORY CASTLE, with overlap with earlier activity (LAZYBONES directly).<br>- BONESAW works against vCenter 7.x and 8.x.</p><p>Reported by Broadcom. The flaw, the affected and fixed versions, no workaround. Not stated by Broadcom, exploitation in the wild.</p><p>Inferred (this analyst, not in the sources). If the signing key was stolen, forged tokens are a problem separate from the flaw, and patching alone does not end them. Google does not say this in so many words.</p><p>Not known.<br>- How many organizations were affected and for how long. SecurityWeek relays Mandiant&#39;s CTO saying multiple incidents, and Google describes one.<br>- What was taken beyond what is listed.<br>- Whether the exfiltrated data reached the actor. Google says &quot;likely&quot;.<br>- Whether other actors now use the flaw.<br>- Whether a public exploit exists.</p><p>In your report: every finding, including negative and inconclusive ones, each with its hypothesis, its result and its evidence. State what is notable in the behavior technically and without hedging. Put unknowns in their own list, as here.</p><p>---</p><p>ATT&amp;CK MAPPING (NOT CHECKED)</p><p>This report: statuses are Reported (by Google or Mandiant). Nothing was observed here. None of these IDs was checked on attack.mitre.org. They are the analyst&#39;s candidate mapping, and Google&#39;s report does not list its own in the page text. Check them before publishing.</p><p>T1190 Exploit Public-Facing Application (BeyondTrust Remote Support, Reported)<br>T1046 Network Service Discovery (nmap scans, Reported)<br>T1505.003 Server Software Component, Web Shell (Reported)<br>T1548.001 Abuse Elevation Control Mechanism, Setuid and Setgid (SUID on /bin/env, Reported)<br>T1136 Create Account (SSO and local administrator accounts, Reported)<br>T1606.002 Forge Web Credentials, SAML Tokens (Reported)<br>T1021.004 Remote Services, SSH (vCenter and ESXi, Reported)<br>T1003.003 OS Credential Dumping, NTDS (files extracted from a guest disk, Reported, fit to check)<br>T1070.004 Indicator Removal, File Deletion (Reported)</p><p>In your report: Technique ID, name and status (Observed, Not Observed, Baseline Behavior), based on what your own evidence shows.</p><p>---</p><p>IOC SECTION</p><p>This report: all from Google and Mandiant. Google also publishes indicators in a GTI Collection for registered users and YARA rules in its post.</p><p>Network. Suspected obfuscation network IP addresses, which Google ties to compromised third party firewalls and routers exposing the SSH banner below on high ports. They are other organizations&#39; devices, so treat a match as a lead, not proof.<br>103.147.251[.]5<br>103.190.38[.]3<br>118.174.128[.]34<br>118.174.128[.]36<br>118.174.128[.]38<br>176.9.89[.]66<br>185.71.15[.]109<br>220.121.128[.]41<br>222.108.77[.]44<br>61.216.56[.]127<br>81.45.89[.]43</p><p>File<br>ActiveSer.jsp, SLIMSAVAGE web shell<br>u.jsp, CAGEMATCH staging web shell<br>/tmp/123, BONESAW<br>/tmp/sss, SAWSNAKE<br>/tmp/dbr, BEARBONES<br>/usr/bin/vmtoolscs, LAZYBONES<br>Also named in the post. /tmp/gdb_file, /tmp/rar, /tmp/123.rar, /tmp/123.key, o.tar.gz, vmfs_browser.py, a hidden directory .a on a VMFS volume, and an authorized_keys entry on ESXi root.</p><p>Behavioral<br>SSH banner SSH-2.0-pulsesecure_23345. Listeners on TCP 53 and 33992 on a Linux appliance. A PUT to /rest/appliance/access/ssh. A local account named sps followed by a redacted string at vsphere.local. A web shell that checks pwd equal to 123. SUID set on /bin/env. nmap -sS with the port lists under Technical Evidence. Slack API calls from a server (LAZYBONES).</p><p>Detections named by Google. YARA rules G_Hunting_Webshell_SLIMSAVAGE_1, G_Hunting_Webshell_CAGEMATCH_1, G_Hunting_Utility_SAWSNAKE_2, G_Hunting_Utility_BONESAW_2, G_Hunting_Backdoor_BEARBONES_1 and G_Hunting_Backdoor_LAZYBONES_1. Google SecOps rules VMware vCenter VAMI API SSH Enablement, Base64 Encoded Shell Command Execution, Linux Web Server Process Echo Base64 Encoded String, Suspicious Curl Network Connection to Staging Directory, Suspicious Curl Download to tmp.</p><p>In your report: Network, File and Behavioral indicators from your own evidence, with a note on what may be shared with partners.</p><p>---</p><p>EXPOSURE NOTES</p><p>This report:<br>- A management plane such as vCenter holds the keys to identity, hosts and virtual machine data. Google calls virtualization infrastructure a prime target and says relying on reactive patching alone is insufficient.<br>- A vCenter patched for the July flaw was attacked through a different one.<br>- The Linux appliances the actor used had no EDR.<br>- Google says the default VCSA host firewall, managed through VAMI, has real limits. It has no port level rules, so trusting an IP often means trusting every port. It is managed through the interface an attacker can reach. It sends no remote logs for denied connections or shell activity. It manages inbound traffic only.<br>- Broadcom lists no workaround, and vCenter 7.0 has no public patch listed.</p><p>In your report: attack surface and risk pathways the hunt surfaced, regardless of outcome.</p><p>---</p><p>OPERATIONAL IMPACT</p><p>This report:<br>- If an attacker held the SAML signing key, they could impersonate users to systems that trust vCenter tokens, and patching alone would not end that.<br>- Per Google, files were read out of a guest virtual machine disk, including NTDS and registry hive files, so directory secrets may be exposed.<br>- The vCenter master key and the PostgreSQL password were taken.<br>- Patching vCenter needs a short loss of management access, and The Hacker News describes an upgrade conflict on one path (VCF 9.1.1).</p><p>In your report: what this means for the business and what it changes, not only what happened.</p><p>---</p><p>TIMELINE</p><p>This report: from Google unless marked, dates as given, times unknown.</p><p>Early 2025. Mandiant sees LAZYBONES in a separate ARMORY CASTLE campaign.<br>Jul 13. BONESAW build timestamp (can be modified).<br>Jul 29. CVE-2026-59309 disclosed. Mandiant has evidence ARMORY CASTLE knew of CVE-2026-123890 before this date.<br>Late August. BeyondTrust zero day exploited, per BleepingComputer only.<br>Early September. ARMORY CASTLE exploits CVE-2026-123890 and moves through vCenter and ESXi.<br>Oct 15. Google publishes an earlier post on the BeyondTrust zero day CVE-2026-102764, patched in BT26-05.<br>Nov 19. Broadcom patch and advisory VMSA-2026-0009. Google publishes this report.</p><p>In your report: a chronological summary in one clock (say which), if the hunt&#39;s length or complexity warrants one.</p><p>---</p><p>DETECTION OPPORTUNITIES</p><p>This report: ideas from the analyst, informed by Google&#39;s own detections. None tested.</p><p>1. Alert on any new SSO administrator account, local administrator account or administrator group change with no matching ticket.<br>2. Alert on a VAMI SSH enable, and on a disable within a short time after, and on SSH logins that follow (Google has a SecOps rule for the enable).<br>3. Flag connections to vCenter&#39;s TCP 2012 from any host that is not a known peer, and any host that scans several VMware ports.<br>4. Send vCenter appliance logs to a remote SIEM, since the local firewall does not log denied connections.<br>5. On Linux appliances, alert on executables and keys in /tmp, SUID changes, gdb use, base64 into files, and outbound Slack calls. Where there is no EDR, close that gap first.<br>6. On ESXi, alert on root SSH, vpxuser Host Client logins from unusual sources, authorized_keys changes, and new hidden directories on VMFS volumes.<br>7. Restrict which hosts can reach vCenter and ESXi management at all, and apply egress filtering, so anomalies stand out. Google recommends this approach and its own hardening tool.<br>8. Scan Linux appliance disks with Google&#39;s YARA rules.<br>9. Treat the remote support appliance as an internet facing system with its own monitoring.</p><p>In your report: specific, buildable improvements based on your findings.</p><p>---</p><p>RECOMMENDATIONS</p><p>This report: R marks Google, Mandiant or Broadcom advice as reported. A marks the analyst&#39;s recommendation.</p><p>Immediate<br>Apply the Broadcom patch. (R, Google and Broadcom)<br>Examine your systems for evidence of compromise and do not stop at patching. (R, Mandiant&#39;s CTO via SecurityWeek)<br>Run the Methodology steps back to early August. (A)<br>Scan Linux appliances with Google&#39;s YARA rules and check the IOC list. (A)<br>Patch the BeyondTrust Remote Support zero day (BT26-05) if you run it. (A)<br>For vCenter 7.0, which has no public patch listed, contact Broadcom Support about extended support and restrict access in the meantime. (A, from the advisory)</p><p>Short Term<br>If anything suspicious is found, treat the SAML signing key, the vCenter master key and the PostgreSQL password as exposed, and follow Broadcom and Google guidance on replacing them. (A, reasoning from the reported theft)<br>Read Mandiant&#39;s guide, vSphere and BRICKSTORM Malware: A Defender&#39;s Guide, and consider its VCSA Hardening and Logging Tool (default deny inbound and outbound, remote syslog), with testing and change control. (R, Google, an interested party, not tested here)<br>Build the account, group and VAMI SSH alerts. (A)</p><p>Long Term<br>Put virtualization management behind a separate administrative network and monitor it as a tier zero asset. (A)<br>Add EDR or equivalent telemetry to Linux appliances, and add vendor appliances with remote access to asset and patch tracking. (A)</p><p>In your report: Immediate, Short Term and Long Term.</p><p>---</p><p>REPEATABILITY NOTES</p><p>This report: another analyst needs the source list with dates, the window and why, the baseline, the IOC and file list above, and the untested logic in Methodology translated to their platform. Recheck every source before running, because details are changing. Google&#39;s page is the source for indicators, and it says the full set is in a GTI Collection for registered users.</p><p>In your report: what another analyst needs to reproduce the hunt without reconstructing your intent.</p><p>---</p><p>APPENDIX AND REFERENCES</p><p>This report: sources, accessed Nov 20, 2026.<br>Google Threat Intelligence Group and Mandiant, ARMORY CASTLE Exploits vSphere, Nov 19, 2026: https://cloud.google.com/blog/topics/threat-intelligence/armory-castle-exploits-vsphere<br>Broadcom advisory VMSA-2026-0009: https://support.broadcom.com/web/ecx/support-content-notification/-/external/content/SecurityAdvisories/0/39495<br>NVD, CVE-2026-123890.<br>BleepingComputer, Lawrence Abrams: https://www.bleepingcomputer.com/news/security/vmware-patches-vcenter-zero-day-exploited-in-attacks-since-september/<br>SecurityWeek, Eduard Kovacs, Nov 19: https://nvd.nist.gov/vuln/detail /CVE-2026-123890<br>The Hacker News, Swati Khandelwal, Nov 19: add the link.</p><p>In your report: list the file names and the naming convention for saved data (exports, CSVs, screenshots) so another analyst can find the work. Example convention: hunt_&lt;source&gt;_&lt;YYYYMMDD&gt;_&lt;description&gt;.csv</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=930fe121e4fa" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[A Threat Hunt Report for the Optimizely Supply Chain Compromise]]></title>
            <link>https://medium.com/@brent.hachey/a-threat-hunt-report-for-the-optimizely-supply-chain-compromise-fd118db89822?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/fd118db89822</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[infosec]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Tue, 17 Nov 2026 14:40:50 GMT</pubDate>
            <atom:updated>2026-11-17T14:40:50.477Z</atom:updated>
            <content:encoded><![CDATA[<p>THREAT HUNT REPORT</p><p>Optimizely supply chain compromise, Nov 2026</p><p>Written by Brent Hachey</p><p>Facts as of Nov 17, 2026, about 14:00 UTC. The incident is still developing, so recheck the sources before publishing.</p><p>HOW TO READ THIS REPORT</p><p>This is a reconstruction built from public reporting. It was not run against any environment. Each section has two parts. &quot;This report&quot; records the facts for this incident and says who observed them. &quot;In your report&quot; says what to write in your own version and where the information comes from. Examples are marked as examples. Format blocks are blank on purpose and contain no real data.</p><p>Status marks. [DRAFT] means written and ready for review. [CONFIRM] means a point the author must confirm.</p><p>---</p><p>HANDLING [DRAFT]</p><p>This report: Public. Built from public reporting. No employer data and no internal telemetry.</p><p>In your report: apply your own classification and retention rules. Example from the original template. Classification: &quot;Confidential&quot;. Handling: &quot;Store in approved internal repositories only. Do not distribute outside Security Operations. Redact hostnames, usernames, and internal identifiers before external sharing. IOC sections may be shared with trusted partners if required. Treat raw logs, screenshots, and credential artifacts as restricted. Follow organizational retention and destruction policy.&quot;</p><p>---</p><p>HUNT DETAILS [DRAFT]</p><p>This report: Prepared Nov 17, 2026. Not run, so no start or end date.</p><p>Lead analyst: Brent Hachey.</p><p>Data sources needed: DNS resolver logs. Web proxy or secure web gateway logs. EDR network and DNS telemetry (Sysmon 3 for network connections and Sysmon 22 for DNS queries, which ties a lookup to the process that made it). Browser content security policy reports. Scheduled snapshots of your own pages&#39; script requests. Optimizely project and change history, if available. Mobile device management records for iOS version. Keep 90 days of history for each where you can.</p><p>In your report: dates, lead and supporting analysts, and the data sources you actually used with the retention you had for each.</p><p>---</p><p>EXECUTIVE SUMMARY [DRAFT]</p><p>This report: On Nov 16, 2026, Report URI found malicious JavaScript served to visitors of 178 organizations through Optimizely Web. The code was added to the optional Project JavaScript field of customer projects, so most files kept arriving from a host every affected site already trusted. Hudson Rock found an earlier wave from Nov 13 through three other script providers, so Optimizely was not the only route. Report URI also found that only 24 of the 159 hosts in its content security policy results (15%) had a policy that blocked the payload. This report turns the public indicators into a hunt for two places, your own Optimizely projects and your users&#39; browsing history from Nov 13 to Nov 17 UTC. It separates what was observed, what is inferred and what is not known. A DNS hit shows a device reached a domain, not that data was taken.</p><p>In your report: three to five sentences on what you examined, what you found and why it matters. Example from the original template: &quot;This hunt assessed workstation to server authentication behavior across a 90 day baseline to determine whether recent privileged transitions represented legitimate workflows or early stage credential misuse. No malicious activity was identified, but several detection and visibility improvements were recommended.&quot;</p><p>---</p><p>OUTCOME STATUS [DRAFT]</p><p>This report: Not run. Nothing was tested against an environment, so nothing is confirmed or refuted.</p><p>In your report: state one result and why. Confirmed means your logs show a user device reaching one of the listed domains between Nov 13 and Nov 17 UTC, or an Optimizely project containing the added loader. Refuted means full coverage of those sources with no matches. Inconclusive means retention or visibility gaps in the data section prevent either answer. No findings is still a finding, so say which result applies.</p><p>---</p><p>USE CASE ALIGNMENT [DRAFT]</p><p>This report: Use Case: third party script compromise. Category: supply chain, client side. ATT&amp;CK: see the ATT&amp;CK Mapping section. The IDs need checking against the current ATT&amp;CK matrix before publishing.</p><p>In your report: the use case name your program uses, its category, and the techniques the hunt tests.</p><p>---</p><p>HYPOTHESIS [DRAFT]</p><p>This report has two hypotheses because the two tests need different data.</p><p>H1, your own sites: If an adversary altered a script your pages load from a vendor, the file your visitors receive will differ from the one you last reviewed, and it will load or contact a domain outside your 90 day baseline. This hunt exists to confirm or rule that out.</p><p>H2, your users: If your users loaded a poisoned script from an affected site between Nov 13 and Nov 17 UTC, your DNS or proxy history will show a lookup or request for one of the listed attacker domains. This hunt exists to confirm or rule that out.</p><p>Core Principle: All malicious activity is anomalous, but not all anomalous activity is malicious. Optimizely snippets also hold customer experiments (per Report URI), so they can change for legitimate reasons and a change alone is not enough.</p><p>In your report: one statement per hypothesis covering the scenario, the purpose and what malicious activity would look like if present.</p><p>---</p><p>INTELLIGENCE AND CONTEXT [DRAFT]</p><p>This report: sources, and how far to trust each.</p><p>1. Report URI, Scott Helme, Nov 16, with a Nov 17 update (&quot;World&#39;s Largest Brands Infected With Malware Via Optimizely&quot;). First party observation. It kept a copy of every infected file with headers, hash and fetch time. It sells JavaScript monitoring, so its detection claims are a vendor view. Its counts cover only the snippets it found in time, and it says it cannot see how many cards, cookies or devices were affected.</p><p>2. Optimizely status page, &quot;Web Experimentation Security Incident&quot;, Nov 17, 01:02 UTC, status Monitoring. It says an unauthorized party exploited a vulnerability and added malicious JavaScript to some customer projects, and that the vulnerability is fixed and the injected code removed. It gives no code level detail. Report URI says Optimizely had not responded to it at publication. [CONFIRM: this was read through a page fetch tool. Open the page yourself and check the wording and when the incident was first posted.]</p><p>3. Hudson Rock with Kirk of Ohey, Nov 17 (&quot;Beyond Optimizely&quot;). Secondary analysis plus recovered files. It states it has not reproduced the later iOS stages or demonstrated compromise of a real iPhone, and that its evidence does not establish four independently explained vendor breaches or identify a single operator. It has not revalidated Report URI&#39;s counts.</p><p>4. Background on the iOS kits. Google Threat Intelligence Group, &quot;Coruna&quot; (March 2026), and the DarkSword reporting that followed.</p><p>5. Press coverage (BleepingComputer, SecurityWeek). Secondary, used here only for Optimizely&#39;s statements.</p><p>In your report: list each source with its date and link, say whether you observed it or read it, and note any limits or interests.</p><p>---</p><p>ENVIRONMENT SCOPE [DRAFT]</p><p>This report:</p><p>In. Web properties you run that load Optimizely Web or other third party scripts. Staff and customer devices, using DNS and proxy history for Nov 13 to Nov 17 UTC. Managed iPhones and iPads. Even if you do not use Optimizely, your staff may have visited affected sites on Nov 16, so the user side applies to everyone.</p><p>Why this window. Hudson Rock&#39;s earliest observation of the earlier wave is Nov 13 (an observation time, not an intrusion start). The first Optimizely injection in Report URI&#39;s timeline is Nov 16, 04:21 UTC. Nov 17 covers clean up and takedown.</p><p>Deliberately out. Card entry inside hosted payment iframes, because Report URI says the script could not reach them. Optimizely&#39;s own systems. Anything you do not run.</p><p>In your report: systems, accounts and segments included, and each exclusion with its reason.</p><p>---</p><p>BASELINE DEFINITION (30 TO 90 DAYS) [DRAFT]</p><p>This report: two baselines, each ending before the window opens (Nov 12 or earlier) so the incident does not contaminate it.</p><p>Pages. For each page that matters, such as login and checkout, the scripts it loads, the file each one served, and the domains each one called. Compare 7, 14, 30 and 90 day views. Record how often each vendor script normally changes. A change in a script that rarely changes carries more weight. Report URI says an Optimizely snippet contains the customer&#39;s experiments, so it can change for legitimate reasons.</p><p>Users. 90 days of DNS and proxy history, so a domain never seen before the window stands out.</p><p>In your report: what normal looks like, set before you evaluate anomalies, and the dates it covers.</p><p>---</p><p>REQUIRED DATA VS. ACTUAL DATA [DRAFT]</p><p>This report: Required. The sources under Hunt Details, with retention covering Nov 13 to Nov 17 UTC, script snapshots from before Nov 16, and visibility of outbound browser traffic. Actual. Not applicable, not run. Before starting, confirm DNS and proxy retention, whether iOS device traffic is visible, whether any pre Nov 16 snapshots exist, and whether Optimizely keeps a change history for Project JavaScript (not verified here). Browsers using encrypted DNS can bypass resolver logs. Without snapshots you can search for the indicators but cannot show what changed.</p><p>In your report: what telemetry the hypothesis needs, what you actually had, the gaps, and your confidence in each. Example from the original template: &quot;All required data sources were available except Script Block Logging on three legacy servers, reducing visibility into module level PowerShell execution. Confidence is high across workstations and application servers, reduced on the three legacy servers.&quot;</p><p>---</p><p>TOOLS AND AI USE [DRAFT, CONFIRM]</p><p>This report: Tools. Public web sources and vendor posts only. No SIEM, EDR or other platform was used because nothing was run. AI. Claude was used to read and compare the source documents, check claims against them, and draft section language from the template and the analyst&#39;s direction. No query was generated or run and no telemetry was interpreted. The lead analyst validated the claims against the sources before publishing. [CONFIRM: this last sentence is true only once the author has completed that check.]</p><p>In your report: list every tool used to run the hunt, then state plainly whether AI factored in, which tool, which task, and that a human validated its output before it reached the findings. If none, say none.</p><p>---</p><p>METHODOLOGY [DRAFT, UNTESTED]</p><p>This report: logic and windows for each step. Nothing here was run. Translate it to your platform and record the actual queries.</p><p>Step 1. Inventory the scripts on in scope pages, with URL, response hash and domains called. Window: 90 days before Nov 13, plus Nov 13 to Nov 17.</p><p>Step 2. Review each Optimizely project, including Project JavaScript and every experiment&#39;s custom code. Look for a script element created with an external source, atob(, eval, and the domains in the IOC section. Report URI notes one loader variant hid the URL in base64, so a domain search alone misses it, and one variant carried the full payload inline and runs through eval if the domain is blocked.</p><p>Step 3. Search DNS and proxy logs, Nov 13 to Nov 17 UTC, for the network indicators. Group by device and user and record first and last seen. The main domain no longer resolves, so the lack of a live answer today says nothing about history.</p><p>Step 4. Where EDR or Sysmon 22 exists, identify the process that made each lookup and confirm it was a browser.</p><p>Step 5. For iOS, check the MDM version against Report URI&#39;s guidance (iOS 18.7.3 or later, or 26.3 or later) and look for the .icu router domains in DNS. Report URI reports that pages on iOS 18.6 or earlier reload every 60 seconds to retry the exploit, which is a visible symptom.</p><p>Step 6. For each hit, determine the pages visited and the times, and whether card entry or crypto transfers happened on Nov 16.</p><p>In your report: the structured process in sequence, with the actual queries and the logic and time window for each. Example from the original template: &quot;Queried EDR platform for process creation events where parent process equals winlogon.exe, 90 day window.&quot;</p><p>---</p><p>TECHNICAL EVIDENCE [DRAFT]</p><p>This report: public artifacts, each credited to who observed it.</p><p>Report URI. Added to Zoom&#39;s Optimizely snippet, inside projectJS, was a function that creates a script element with the source https://1688360[.]cc/track.js and appends it to the page head, or to the document element if there is no head. Report URI&#39;s timeline dates the Zoom snippet change to 11:20 UTC on Nov 16 from its Last-Modified header.</p><p>Report URI. Three loader versions. A plain script tag to the domain. The same URL hidden in base64 and decoded with atob(. A full inline copy of the payload that runs through eval if the domain is blocked.</p><p>Hudson Rock. Directly recovered files, eight Optimizely project files and the 561 byte minute-ly loader. The Kissmetrics and StorifyMe responses come from archived captures and were not recovered directly.</p><p>Hashes are in the IOC section.</p><p>In your report: raw artifacts such as event IDs, command lines, log excerpts and ticket patterns. Format only, no real data: Source | Timestamp UTC | Host or user | Process | Domain or URL | Result.</p><p>---</p><p>BEHAVIORAL SUMMARY [DRAFT]</p><p>This report: Per Report URI the payload (track.js) was a multi purpose kit with every part active at once. It swapped cryptocurrency wallet addresses in page text, forms, links, the clipboard and outgoing requests. It watched form submissions and requests for a card number, expiry and CVV and sent them to the attacker&#39;s server (the skimmer was added at 13:43 UTC, and cards in hosted payment iframes were out of reach). When it swapped an address near wallet or payment wording it sent the page URL, cookies not marked HttpOnly, the original addresses and a screenshot rendered with html2canvas. On iPhones and iPads it loaded a boot script from two .icu domains, read the iOS version and loaded an exploit chain in a hidden iframe. That was Coruna for iOS 13 to 17, DarkSword for 18.x, and a separate chain for iOS 26 that Report URI says only fingerprinted the device. The payload was rebuilt 26 times on Nov 16.</p><p>In your report: what your environment did, synthesized from the evidence.</p><p>---</p><p>FINDINGS [DRAFT]</p><p>This report: no findings, because nothing was run. The public record separates into four lists.</p><p>Observed by Report URI. Code added to Project JavaScript in 188 snippets at 178 organizations. Verified transactions to the attacker&#39;s wallets, about $7,040 as of 19:25 UTC on Nov 16 and about $7,150 by adding the wallet table in the Nov 17 update (Report URI gives no total there). Code capable of capturing cards, cookies and screenshots. iOS loaders serving a chain chosen by version.</p><p>Observed by Hudson Rock. Poisoned files from Kissmetrics, StorifyMe and minute-ly, with first observations on Nov 13 (securitybank.com and faz.net) and Nov 16 (foxsports.com). webshell[.]lol/get.js as a shared loader. The iOS router pages responding to a scan with an iPhone user agent (a delivery observation, not an exploitation test).</p><p>Inferred. Report URI ties the actor behind this wave to earlier thefts through a cash out address reused from earlier operations.</p><p>Not known. How the attacker got write access across unrelated customer projects (Optimizely says a vulnerability was exploited, with no code level detail). How the other three providers were reached. How many cards, cookies, visitors or devices were affected. Whether any real iPhone was compromised (Hudson Rock has not demonstrated one). Whether the wallet total reflects loss from this incident (Hudson Rock says it has not linked the transfers to affected visitors).</p><p>In your report: what is notable in the behavior, stated technically and without hedging. Put unknowns in their own list, as here.</p><p>---</p><p>ATT&amp;CK MAPPING [DRAFT, PARTLY VERIFIED]</p><p>This report: the original template&#39;s statuses are Observed, Not Observed and Baseline Behavior. Nothing was observed here, so this report uses Reported (by Report URI or Hudson Rock).</p><p>T1195 Supply Chain Compromise (Reported)</p><p>T1189 Drive by Compromise (Reported)</p><p>T1059.007 Command and Scripting Interpreter, JavaScript (Reported)</p><p>T1027 Obfuscated Files or Information, base64 loader (Reported)</p><p>T1539 Steal Web Session Cookie, cookies not marked HttpOnly (Reported)</p><p>T1113 Screen Capture, html2canvas screenshot (Reported, weak fit, because ATT&amp;CK describes desktop screen capture and this is a page render)</p><p>T1565.002 Transmitted Data Manipulation, address swapping in requests and the clipboard (Reported, MITRE lists a clipboard address swapper under this technique)</p><p>T1195, T1189, T1539 and T1565.002 were checked on attack.mitre.org on Nov 17. T1059.007 and T1027 have not been checked. Card capture needs a technique chosen after checking the matrix.</p><p>In your report: Technique ID, name and status (Observed, Not Observed, Baseline Behavior), based on what your own evidence shows.</p><p>---</p><p>IOC SECTION [DRAFT]</p><p>This report: all from Report URI and Hudson Rock, defanged. Attacker owned per Report URI. Block including subdomains.</p><p>Network</p><p>1688360[.]cc, payload and collection (track.js, card.php, hit.php)</p><p>5vobauua3ojpklx[.]icu, iOS router (boot.js)</p><p>eil786npto9u5bq[.]icu, iOS router</p><p>l7bcsnwc7nzr3uo[.]icu, Coruna telemetry (POST /api/iptj)</p><p>webshell[.]lol, injector (get.js), earlier wave</p><p>linkforge[.]cc, SEO panel script carrying the injector</p><p>je4ft3gn8vxudum[.]icu and q1ls1ho9x7wsxlu[.]icu, fallback servers hard coded in boot.js</p><p>ios-hack[.]icu, fallback (Hudson Rock saw an attempt only, no completed response)</p><p>p55abnw2pnsz0bo[.]icu, 1y69gy2mdzaltz1[.]icu, gm38duxoihbmy5o[.]icu, same certificates and IPs as the kit</p><p>IP addresses are listed in Report URI&#39;s post. They can be shared hosting, so treat an IP match alone as weak.</p><p>Status as of Report URI&#39;s Nov 17, 09:17 UTC update. 1688360[.]cc on clientHold since 00:37 UTC and no longer resolves. webshell[.]lol on clientHold since 03:37 UTC. eil786npto9u5bq[.]icu no longer resolves. 5vobauua3ojpklx[.]icu resolves but does not respond.</p><p>File (SHA 256)</p><p>track.js, 17:13 build: 814a963c0da480e2449d9b6664480156e1cd6d731014117401bf635badec3a0f</p><p>track.js, latest build 17:57, 29,882 bytes: f7b84b9a94f749f8ced3f0112d0a66894e8bbc3b4aa5592021d1b8976a9bad40</p><p>Decoded inline fallback, 24,392 bytes (same in the BBC and HSBC UAE files, per Hudson Rock): c2a34c579e4c2e8802752c93b3a9cca0230610f3e4dd6688f4272459c7c7c162</p><p>Kissmetrics i.js response, 213 bytes: c154f3b644cd51f2446483c7ffd2ae4f178c4b0cb54846cfb5b83215367349cf</p><p>StorifyMe widget response, 16,569 bytes: 94d174abe1e3b12b000f2bc8cbd8e9344290f4b5a8661c8d700cdc5675675fba</p><p>minute-ly loader.js, 561 bytes: 0f2f2a27d19112e5800541daecc91eb6c77e49f9b43ed23c890b876631978533</p><p>BBC, earlier archived version: 45e42f5fa7520f6e7c048587f025549b31f0f24be8be058c7037d1da37e37b67</p><p>BBC, later file: 305d0b00a7bafae9f240dbaf5bfdbb19648e38e509e8992112fcb7866c2e37dc</p><p>The two BBC files are different versions, so do not assign the later file&#39;s fallback capability to the earlier one (Hudson Rock).</p><p>Behavioral</p><p>Global markers window._k7 and window.__k7b. Script attribute data-channel-code=&quot;02111110&quot;. Telemetry channel code BIGNIAO. A loader that hides its URL with atob(. An inline copy that runs through eval. Pages on iOS 18.6 or earlier reloading every 60 seconds. Attacker wallet addresses (BTC, ETH and EVM, TRON, SOL, TON, LTC, BCH, DOGE) are in Report URI&#39;s post.</p><p>In your report: Network, File and Behavioral indicators from your own evidence, with a note on what may be shared with partners.</p><p>---</p><p>EXPOSURE NOTES [DRAFT]</p><p>This report: Third party scripts and vendor projects with a custom JavaScript field can change what runs on your pages outside your change control. On content security policy, Report URI crawled 256 of the affected organizations&#39; hosts. Its results table covers 159 of them and does not explain the other 97. Of those 159, 68 had no policy, 50 allowed the payload and the data theft, 17 ran the payload but connect-src blocked the data theft (address swapping still worked), and 24 (15%) blocked the payload. &#39;strict-dynamic&#39; and &#39;unsafe-eval&#39; let it through, and Report URI says Optimizely&#39;s own documentation tells customers to allow &#39;unsafe-eval&#39;. Allowing the vendor&#39;s host did not help because the code came from that host.</p><p>In your report: attack surface and risk pathways the hunt surfaced, regardless of outcome.</p><p>---</p><p>OPERATIONAL IMPACT [DRAFT]</p><p>This report: If you use Optimizely Web, Report URI advises treating cards typed into your own forms on Nov 16, and session tokens or secrets in cookies not marked HttpOnly, as potentially exposed, and considering whether to tell your customers. If you do not, staff who visited affected sites on Nov 16 may have pasted a swapped crypto address or paid by card on an affected site, and iPhones or iPads on older iOS may have been served an exploit kit (delivery observed, exploitation not demonstrated).</p><p>In your report: what this means for the business and what it changes, not only what happened.</p><p>---</p><p>TIMELINE [DRAFT]</p><p>This report: UTC, from Report URI unless marked.</p><p>Sep 17. webshell.lol, the same actor&#39;s injector, first appears on Turkish sites through an SEO backlink panel script.</p><p>Nov 9, 17:52. 5vobauua3ojpklx.icu registered (iOS exploit kit router).</p><p>Nov 13. Earlier wave. Hudson Rock&#39;s first observations are Kissmetrics on securitybank.com at 05:42 and StorifyMe on faz.net at 20:31 (observation times, not intrusion starts).</p><p>Nov 16, 02:21. Last clean urlscan capture of the BBC&#39;s Optimizely snippet.</p><p>Nov 16, 04:21. First Optimizely injection seen (BBC snippet loading webshell.lol).</p><p>Nov 16, 04:28 to 05:23. 1688360.cc certificates issued. Domain registered at 05:23.</p><p>Nov 16, 08:10. First build of track.js (crypto clipper).</p><p>Nov 16, 08:28. 1688360.cc/track.js first seen on bbc.com.</p><p>Nov 16, 11:20. Zoom&#39;s snippet modified (its Last-Modified header).</p><p>Nov 16, 13:03. Attacker sweeps stolen crypto to exchanges.</p><p>Nov 16, 13:43. Card skimmer added to track.js.</p><p>Nov 16, 16:38. eil786npto9u5bq.icu registered (second iOS router).</p><p>Nov 16, 17:13 to 17:57. Second iOS host added. Coruna iframe injected into every iOS 18.6 or earlier visit.</p><p>Nov 16, 19:24 to 19:59. Report URI sweep, 186 of 1,611 snippets infected.</p><p>Nov 16, about 20:40 to 21:10. Mass clean up.</p><p>Nov 16, 21:52. 1 of 188 snippets still infected (TalkTalk).</p><p>Nov 16, 23:11. All infected Optimizely files cleaned.</p><p>Nov 17, 00:37. 1688360.cc put on clientHold.</p><p>Nov 17, 01:02. Optimizely status page update, status Monitoring.</p><p>Nov 17, 03:37. webshell.lol put on clientHold.</p><p>Nov 17, 04:11 to 05:39. Attacker empties the wallets.</p><p>In your report: a chronological summary in one clock (say which), if the hunt&#39;s length or complexity warrants one.</p><p>---</p><p>DETECTION OPPORTUNITIES [DRAFT]</p><p>This report: ideas, none tested. These are examples to get you thinking about your environment.</p><p>1. Alert when a vendor script changes and calls a domain it has not called before. A change alone is noisy.</p><p>2. Hash and store every third party script on key pages, keep the files, and diff them.</p><p>3. A content security policy in report only mode shows new script sources and destinations from real browsers. Restrict connect-src.</p><p>4. Rank new domains by age, TLD, how random the name looks, and whether it matches the loading vendor&#39;s known domains. Here 1688360.cc was registered about three hours before its first sighting on bbc.com, and the .icu router names were random strings. Each signal alone is weak.</p><p>5. Search stored scripts for atob( and eval as well as domains.</p><p>6. Crawl with desktop and iPhone profiles, because the payload chose its behavior by iOS version. Crawl hourly, because Report URI&#39;s timeline shows roughly 16 hours between the first injection and the mass clean up, and a daily crawl could miss it.</p><p>7. Where process level DNS exists, join lookups to the browser process so a hit can be tied to a user session.</p><p>In your report: specific, buildable improvements based on your findings.</p><p>---</p><p>RECOMMENDATIONS [DRAFT]</p><p>This report: R marks Report URI&#39;s advice. A marks the analyst&#39;s recommendation.</p><p>Immediate</p><p>Open every Optimizely project&#39;s Project JavaScript and every experiment&#39;s custom code. Look for 1688360[.]cc, atob( and any script you did not write. (R)</p><p>Review who and what has write access to the Optimizely account (users, API tokens, integrations) and rotate credentials. (R)</p><p>Search DNS and proxy history for Nov 13 to Nov 17 UTC for the network indicators. (A)</p><p>Treat cards typed into your own forms on Nov 16, and session tokens in cookies not marked HttpOnly, as potentially exposed. Rotate the tokens. (R)</p><p>Update iPhones and iPads to iOS 18.7.3 or later, or 26.3 or later. (R)</p><p>Short Term</p><p>Deploy a content security policy, restrict connect-src, avoid &#39;unsafe-eval&#39; where you can, and turn on reporting. (R) Start in report only mode so nothing breaks, then enforce. (A)</p><p>Build the page script baseline and the change plus new domain alert. (A)</p><p>Long Term</p><p>Put vendor script configuration under change control with review and alerting. (A)</p><p>Add integrity monitoring and vendor access review to onboarding for third party scripts. (A)</p><p>In your report: Immediate, Short Term and Long Term.</p><p>---</p><p>REPEATABILITY NOTES [DRAFT]</p><p>This report: another analyst needs the source list with dates, the window (Nov 13 to Nov 17 UTC) and why, the two baselines, the IOC list with sources, and the untested logic in Methodology translated to their platform. Recheck every source before running, because the indicators and takedown status are changing.</p><p>In your report: what another analyst needs to reproduce the hunt without reconstructing your intent.</p><p>---</p><p>APPENDIX AND REFERENCES [DRAFT]</p><p>This report: sources, accessed Nov 17, 2026.</p><p>Report URI: https://blog.report-uri.com/worlds-largest-brands-infected-with-malware-via-optimizely/</p><p>Optimizely status: https://status.optimizely.com/</p><p>Hudson Rock: https://www.infostealers.com/article/beyond-optimizely-malicious-scripts-traced-across-four-web-services/</p><p>Google Threat Intelligence Group, Coruna: https://cloud.google.com/blog/topics/threat-intelligence/coruna-powerful-ios-exploit-kit</p><p>SecurityWeek, Eduard Kovacs, Nov 17: https://www.securityweek.com/optimizely-vulnerability-exploited-to-serve-malware-on-major-websites/</p><p>BleepingComputer, Lawrence Abrams: add the link.</p><p>In your report: list the file names and the naming convention for saved data (exports, CSVs, screenshots) so another analyst can find the work. Example convention: hunt_&lt;source&gt;_&lt;YYYYMMDD&gt;_&lt;description&gt;.csv</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=fd118db89822" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part Three — Where a Hunt Actually Takes Shape]]></title>
            <link>https://medium.com/@brent.hachey/part-three-where-a-hunt-actually-takes-shape-388ed93a2085?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/388ed93a2085</guid>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Wed, 09 Sep 2026 13:30:00 GMT</pubDate>
            <atom:updated>2026-09-09T13:30:00.630Z</atom:updated>
            <content:encoded><![CDATA[<p>Part Three — Where a Hunt Actually Takes Shape</p><p>Part Two handed off at the exact moment the data confirmed the hypothesis can actually be tested. The preparation is done. The scope is unlocked. What happens next is where a hunt either holds its discipline or quietly loses it.</p><p>This is Narrowing Your Analysis, the phase most people picture when they hear the word threat hunting. The reality is that without everything that came before it, analysis has nothing to anchor to. You are no longer asking whether the hunt is possible. You are measuring whether the behavior you predicted actually exists in the environment, and that question demands something most analysts underestimate. It demands the willingness to be wrong in front of your own documentation without quietly reshaping the hypothesis to fit whatever you happened to find.</p><p>WMI abuse is a good example of what that narrowing actually looks like in practice. I check for it constantly, in incident response daily and in a large share of the hunts I run when the hypothesis calls for it. WMI activity is inherently noisy. Windows leans on it constantly for legitimate system administration, which is exactly why it matters when it gets leveraged for persistence or privilege escalation instead.</p><p>Attackers favor it specifically because commands run through it execute under the WmiPrvSE.exe host process. That means they inherit elevated privileges rather than the restricted session the attacker actually started in. That single fact is why the data matters.</p><p>A command that would normally run as a limited local user instead shows up running as SYSTEM. The only way to catch that is by watching what WmiPrvSE.exe actually does. Sysmon Event ID 19 shows a WMI event filter getting created. Event ID 20 shows an event consumer tied to it. Event ID 21 shows the two getting bound together, the actual persistence mechanism taking shape. None of those three events alone proves anything. What you are actually looking for is the moment WmiPrvSE.exe spawns something it has no legitimate reason to spawn. A shell process. A registry modification. A user account change. That parent to child relationship is the signal underneath the noise. That is narrowing. Individually unremarkable events correlated into one clear picture of persistence or escalation nobody would catch by looking at any single log line alone.</p><p>This is also where the mind starts working against you if you let it. Picture the moment itself. Thirty minutes into a hunt, a process spawn shows up that is close to what your hypothesis predicted but not quite an exact match. The temptation is to round it up to a confirmation because you already spent thirty minutes getting there. That drift rarely feels like drift while it is happening. It feels like instinct. It feels like experience. The two are not the same thing, and confusing them is how a structured hunt collapses back into unstructured searching.</p><p>The real discipline of Narrowing Your Analysis comes down to a distinction most analysts never learn to hold apart cleanly. A fact is what the data shows, stated in terms that do not require anyone&#39;s confidence rating to be true. A process spawned a child process it has no documented reason to spawn. A credential authenticated from a location it has never authenticated from before. Those are facts, verifiable and reproducible by anyone who pulls the same data. A judgment is the meaning you assign to that fact, the story you build around why it happened and what it means for the hypothesis you are testing. Judgments are necessary. Every hunt eventually needs them. But the moment a judgment gets treated with the same certainty as a fact the entire chain of reasoning behind it inherits a confidence it never actually earned. That is not a technical error. It is a cognitive one. It mistakes a clean narrative for proof.</p><p>A malicious WMI artifact is never the whole story. In incident response I see this almost every day. The subscription or the rogue process spawn is a bridge between attack phases, not the attack itself. Finding it is where the pivot starts. If the artifact is a permanent event subscription running an encoded payload, the next question is what infrastructure that payload is calling out to. Pull the destination from the command line and pivot to firewall, proxy, and DNS logs to find out when that communication started and whether any other hosts are already talking to the same address. Then pivot backward in time on the same host. WMI subscriptions do not create themselves. The exact timestamp of subscription creation is visible in Sysmon Event IDs 19, 20, and 21. That timestamp becomes a window. Search everything that happened on that host in the ten to fifteen minutes before it and the original access vector starts to surface. If the artifact is WmiPrvSE.exe spawning a shell process, the pivot goes to authentication logs instead. A Type 3 network logon at the exact second that process spawned gives you the source IP and the account that triggered the WMI execution remotely. That source host becomes the next machine you investigate, and that is usually where patient zero is waiting. The more hunts you run the sharper that instinct gets. You stop seeing individual events and start reading the chain they belong to.</p><p>There is a moment in every hunt where the pivoting has to stop. Not because you have found everything, but because the chain has given you enough to either prove or disprove the hypothesis. Experienced analysts feel it before they can explain it. The pivots start returning familiar ground. The same systems and accounts and timestamps start circling back on themselves. That is the signal. The exploration phase is over. What you have collected either builds into a case or it does not. Switching gears at that moment from chasing to building is one of the hardest transitions in the discipline. The instinct is to keep pulling threads. The discipline is to recognize when the threads are ready to be woven into something defensible.</p><p>As the hunt progresses the investigation shifts from exploring possibilities to collecting evidence. Evidence carries a psychological weight most people do not expect until they are the one responsible for defending it. Evidence is not a single log line or a screenshot pulled in the moment. It is a chain of behavior that has to be attributable, repeatable, and preserved with enough context that someone who was not in the room can walk through your reasoning and land on the same conclusion. That requirement exists specifically to protect the hunt from its own analyst. It forces you to build a case sturdy enough to survive someone else&#39;s skepticism and not just your own confidence.</p><p>Evidence also has a way of exposing what is not there. When the data a hypothesis needs simply does not exist that absence becomes part of the finding. It tells you exactly where the organization is blind and what risk currently cannot be measured at all. A hunt that surfaces a visibility gap and nothing else has still produced something real. Proof of a blind spot is proof. It just is not the proof most people were hoping to walk away with.</p><p>Every piece of evidence you collect should be answering a question the hypothesis asked. At some point the chain is constructed and the picture is complete. Continuing to dig past that point is not thoroughness. It is aimlessness wearing the appearance of discipline. The hypothesis is what keeps you straight. It defines what is relevant and what is noise. On the malicious side, you do not stop collecting until the attack chain is fully accounted for. On the legitimate side, you do not document every clean log line you touched along the way. Collecting irrelevant evidence does not strengthen a report. It overwhelms the reader and buries the actual finding underneath material that was never part of the case. This is the part of the hunt that most junior analysts struggle with because it requires committing to what matters and letting go of what does not. Finding evidence feels like progress. Interpreting it and deciding what belongs in the report feels like exposure. The willingness to sit with that discomfort and let the hypothesis define the boundaries is the psychological center of this entire discipline.</p><p>Following the chain only gets you so far. At some point the evidence either builds into something defensible or it does not. The data either shows the behavior you predicted or it shows you something else entirely, and neither outcome means the hunt failed. What matters is whether you followed the evidence instead of the answer you were hoping for. Analysts who have run enough hunts know the feeling of a hypothesis that is not landing the way they expected. The instinct is to push harder in the same direction. The discipline is to stop, reassess, and let the data reframe the question instead.</p><p>Once the conclusion is reached the work shifts again. Not from one technical task to another but from analyst mode to communicator mode. These are genuinely different cognitive states. In analyst mode the goal is truth. In communicator mode the goal is clarity. The same finding that took hours to surface has to be compressed into language that means something to people who were not in the room when it happened. That compression is where most hunt reports fail. Not because the analyst did the work wrong, but because they never fully made the shift.</p><p>This is Translating Your Findings. Audience is the first challenge. Detection engineering, incident response, and leadership all need the same findings translated differently. A detection engineer needs the exact behavior, the process lineage, the event IDs, and the query logic that would catch it again. An incident responder needs the timeline, the scope, and what to contain first. Leadership needs the business risk framed in plain language, the scope of exposure and the gaps that made it possible. The hunter&#39;s job is to provide those findings clearly and honestly. What leadership does with them from there belongs to a different seat. A well structured report addresses all three audiences in the same document. Each section serves a different reader without requiring a different report.</p><p>Skipping this phase is just as costly. When Translate gets rushed the hunt closes. A ticket gets written and nothing changes. The technique the hunt was built around stays undetected. The gap that made it possible stays open. Six months later the same hypothesis gets run again by someone who never knew it had already been investigated. That is not a hypothetical. It happens constantly in environments where hunting is treated as an event rather than a cycle.</p><p>Before that report gets filed, one more thing has to happen. A second analyst has to look at the facts and the judgments built on top of them and ask whether the reasoning actually holds, independent of how confident the original analyst sounds. Sometimes that means a colleague reads your writeup and asks the one question you were quietly hoping nobody would ask. That is the point of the review. It is not a formality tacked on at the end. It is a second set of eyes with no investment in your conclusion holding up. No single analyst is immune to reading their own conclusion back into their own evidence no matter how sharp they are. Peer review is the institutional version of the discipline every analyst is supposed to be practicing alone.</p><p>Loop closure is what separates a mature program from a collection of one off investigations. When a finding becomes a detection rule it changes what the next hunt needs to look for. Documented gaps become the basis for the next scoping conversation. Escalated blind spots become the intelligence that shapes the next hypothesis. Translating your findings is not the end of the hunt. It is the input that starts the next one. That is what makes HUNT a cycle rather than four steps in a row.</p><p>The artifact that comes out of this phase is the report itself. A threat hunt report is a specialized documentation of potentially exposed risks and vulnerable areas within your network and assets. Some of the best findings have nothing to do with finding an adversary. They are about finding process gaps like asset management failures, ownership blind spots, and areas where accountability was assumed but never confirmed. We all read about ransomware and cyber espionage and perhaps some of us have fantasized about catching one. The most effective threat hunting is often quieter than that. It is closing the gaps before those scenarios ever get the chance to materialize.</p><p>A hunt reaches its natural end when understanding replaces uncertainty rather than papering over it. The value of a hunt was never measured by whether it caught something. It is measured by whether the environment is better understood on the other side of it than it was walking in. Done right, a hunt closes with clarity instead of speculation built on a chain of facts an analyst was disciplined enough to keep separate from what they merely believed. That discipline more than any tool or framework is the entire reason the work holds up under pressure.</p><p>This series covered one complete cycle of a threat hunt across three parts. Part One introduced the HUNT framework and the foundation every hunt depends on before a single query gets written. Part Two confirmed whether that foundation can actually hold weight against real data. Part Three put it to work, narrowing the signal from the noise and collecting what actually matters and translating it into something the organization can act on.</p><p>What this series could not fully carry is the psychological weight that sits underneath all of it. Fear shapes a junior analyst&#39;s first no findings result. Fatigue slowly erodes discipline across a long career. Doubt quietly warps how you read evidence you should be trusting. The framework handles the structure. Only you can handle those.</p><p>A hunt is only as strong as the analyst running it. The structure exists to protect you from your own blind spots. The discipline exists to keep you honest when the pressure to close something starts outweighing the evidence to close it with. That is the work. Not just the queries and the pivots and the reports. The choice to run the process anyway every single time even when nothing comes back.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=388ed93a2085" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[Part Two — Where a Hunt Actually Begins]]></title>
            <link>https://medium.com/@brent.hachey/part-two-where-a-hunt-actually-begins-8106df680d3f?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/8106df680d3f</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Wed, 05 Aug 2026 02:29:08 GMT</pubDate>
            <atom:updated>2026-09-09T13:20:30.159Z</atom:updated>
            <content:encoded><![CDATA[<p>Part Two — Where a Hunt Actually Begins</p><p>When I first began learning how to hunt, I assumed the work started with data. Most analysts do. You open a SIEM, write a few queries, and begin scanning for anything that looks unusual. It took time and experience to understand that a hunt does not begin with searching. It begins with clarity. Before any analysis happens, you need to understand what you are examining, why it matters, and what threat scenario you are trying to confirm or rule out. Without that foundation, the rest of the process becomes unfocused.</p><p>This early phase gets overlooked constantly because it feels less technical than pulling logs, but it is the piece that decides whether the entire hunt means anything. This is where HUNT actually starts, the four part method this whole book walks through, the same idea the subtitle points to before you ever open past the cover. H stands for Hypothesis. Like many analysts early in their career, I thought threat hunting was the coolest title in cybersecurity. I didn&#39;t know the formal structure associated with a hypothesis. Once I learned a little more about it, I understood what it was but not the importance behind the foundation of it. I would run what I thought was a hunt and document my hypothesis at the end, alongside my methodology. What I learned over time is that the hypothesis is what gives the entire hunt its structure. It defines what you&#39;re testing, what data you need to test it, and which outcomes actually matter when the hunt is over.</p><p>Building a hypothesis sounds straightforward until you are actually sitting in the seat doing it. The type of hunt and the intelligence driving it will shape the direction, but the environment you work in will shape the pressure around it. Organizations that fund threat hunting programs want to see metrics and value come back out. Threat hunting is a black box to many professionals and leadership teams, and that misunderstanding creates its own kind of stress. We cover those pressures throughout this book. What matters here is this. When you sit down to build a hypothesis, keep the information true and the scope accurate. Do not start with a conclusion. Start with an idea or a goal intended to shed light in the darkness.</p><p>This phase is Unlocking Your Scope. This is the point where you decide whether your intelligence and hypothesis can actually be tested against the data streams available to you. It is easy to assume the data you need is sitting there waiting for you. Environments rarely offer that kind of clean visibility. You need to know what logs actually exist, how far back retention goes, and what fields those logs actually populate. Whether any of it gives you enough context to validate or kill your hypothesis is a separate question entirely. This is also where the real gaps surface. If the telemetry does not exist, that absence is not a dead end. It is a finding in its own right. It tells you exactly where the organization is blind and what risk currently cannot be measured at all.</p><p>Unlocking your scope means owning the process in the moment. Your job is to confirm you have enough IOCs and the right data sources to match the intelligence driving the hunt. This is also where you will face leadership or customers directly. Both can put high pressure on you to deliver answers they may have already formed in their own minds. Being able to explain what is required and why while holding to the threat hunt policies in place is not optional. Third parties do not dictate the scope of a threat hunt. This is a skill that junior analysts struggle to develop out of fear of how it reflects on their personal brand and employment security. Many organizations have C level executives who push analysts to bend policies and rules under pressure. This is where clarity gets blurred and confusion sets in. Understanding who actually holds the authority to authorize a temporary policy bypass and under what conditions is part of the psychological discipline this work demands. Every exception needs a paper trail. Every deviation needs to be documented for internal use before it becomes a precedent nobody remembers approving. Starting a hunt with missing information only adds time and stress to a process that already demands both. We cannot spend unlimited time searching every system and service and user and action in an environment. Time requires funding and efficiency is what makes threat hunting an easier sell to leadership the next time around.</p><p>Unlocking scope is about feasibility. It confirms the hunt can actually produce evidence instead of assumptions or emotional bias reporting back whatever you wanted to see. This step alone can stop a hunt cold before it ever gets moving. Picture a hypothesis built around anomalous sign in activity. In a SOC or MDR environment, one customer&#39;s Microsoft 365 sign in logs might simply not be ingested, either because an integration was never finished configuring or because the financial commitment does not match the scoped threat hunt request. There is nothing left to hunt at that point. The right move is not to force the hunt anyway. It is to tell the customer directly, explain the exact risk that gap leaves them carrying, and be honest that this particular hunt is hollow or out of scope until the logging exists to support it.</p><p>Part Two is where the foundation gets built. The hypothesis gives the hunt its structure. Unlocking your scope confirms whether that structure can actually be tested. Without both in place, everything that follows runs on assumption instead of evidence. Part Three is where that foundation gets put to work. It is where the data either confirms what you expected or shows you something you did not. That is where a hunt actually takes shape.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=8106df680d3f" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[
What a threat hunt Actually Is

A Threat Analyst’s perspective on the real structure behind threat hunting]]></title>
            <link>https://medium.com/@brent.hachey/how-many-of-us-feel-on-the-daily-e6be7d19719a?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/e6be7d19719a</guid>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <category><![CDATA[education]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Tue, 16 Jun 2026 12:24:04 GMT</pubDate>
            <atom:updated>2026-09-09T13:20:06.812Z</atom:updated>
            <content:encoded><![CDATA[<p><strong>What a threat hunt Actually Is</strong></p><p>A Threat Analyst’s perspective on the real structure behind threat hunting</p><p>When I first started digging into threat hunting, I realized quickly that the term was being used in ways that didn’t match the actual discipline. Teams wanted to hunt, but the definition shifted depending on who you asked. That pushed me to step back and build a clearer understanding of what threat hunting really is and what it requires. What follows is the foundation that shaped how I approach the work in real environments.</p><p>Threat hunting is one of the most misunderstood areas in security. People often describe it as searching through logs or running queries to see if anything suspicious appears. While those actions can happen during a hunt, they are not the core of the discipline. Hunting is not an unstructured exploration of data. It is not a broad search for anomalies. It is not a substitute for detection engineering or alert triage. Threat hunting is a methodical, intelligence driven process with defined objectives, specific data requirements, and a structured workflow. That clarity is what most teams are missing when they say they want to hunt.</p><p>In a previous role, I saw firsthand how difficult it can be for a team to align on what threat hunting actually means. The interest was there. Analysts genuinely wanted to hunt. The definition was the problem. We held internal discussions to agree on terminology, and different analysts referenced different sources to justify their interpretation. This was not a reflection of capability. The team included some of the smartest people I have worked with. The issue was that the team’s responsibilities were broad, and the work structure did not support developing deeper skills in incident response or threat hunting. Analysts were responsible for their own programs in various analyst roles, daily investigations, and a wide range of operational tasks. With that workload, there was little time to run structured hunts or study the frameworks that define them. There was also no clear ownership of the incident response space, which made it difficult to build a repeatable, intelligence led hunting process.</p><p>I also recognized early on that the real gap was not tooling or visibility. It was the specialized skill set required for threat hunting. I developed a genuine interest in finding what was not meant to be found. The unwanted activity. The things hiding in the dark. The behaviors users might not realize they were leaving behind. At first, the idea of a threat hunt sounded exciting in a broad, almost cinematic way. But like many analysts encountering the discipline for the first time, the reality was very different from what I expected. The structure behind a hunt, the documentation, the hypothesis, the intelligence requirements, the indicators that drive direction, all of it was far more rigorous than I initially understood.</p><p>You can study and read theory, but until you change your mindset and your approach, you will not build an effective threat hunting program.</p><p>One of the most important lessons I learned was that a hunt is not defined by what you find. It is defined by the proof you can show about what you did, what you tested, and what you ruled out. That is the part analysts often miss when they move from reactive investigations into proactive hunting. Customers and executives do not just need to know what you discovered. They need to know what you examined, what ran, what did not run, and why the outcome is trustworthy. A successful hunt is not measured by uncovering malware or catching an active intrusion. A successful hunt is one where the hypothesis is tested thoroughly, the intelligence and indicators are followed with discipline, and the documentation clearly shows the reasoning and the results. No findings is still a finding, and in many cases it is the one that matters most. It tells the organization that a specific threat scenario was examined with intent, evidence, and structure.</p><p>A definition that helped anchor my understanding is that threat hunting is the methodical, use case driven, proactive identification of cyber threats within a computing environment or infrastructure. Each part of this definition carries weight. Hunting is methodical, meaning it follows a repeatable process rather than relying on intuition. It is use case driven, meaning it begins with a defined hypothesis or scenario instead of an open ended search. It is proactive, meaning it occurs without waiting for alerts or detections. It focuses on identification, meaning the goal is to confirm or refute a specific idea about potential malicious activity. It is environment specific, meaning hunts are tailored to the systems, data, and threat landscape of the organization.</p><p>Intelligence is the foundation of effective threat hunting. It is not an optional enhancement. It shapes the direction of the hunt and reduces unnecessary noise. Past incidents, internal and external intelligence, knowledge of attacker techniques, and awareness of the environment’s systems, controls, and gaps all matter. Intelligence ensures that hunts are aligned with realistic threats rather than hypothetical scenarios. It makes the process repeatable, defensible, and grounded in evidence.</p><p>Threat hunting exists because adversaries routinely bypass detections, controls fail, and alerts do not catch everything. This is why the discipline has to be proactive rather than reactive.</p><p>After gaining enough structure to understand what a real hunt should look like, I began shaping a program that aligned with these principles. A coworker and I built a case for the direction we wanted to take. The team was extremely busy, so I took ownership of the program on my own initiative. I created reference sheets, documentation, and a threat hunting template that anyone on the team could use. The goal was to build a repeatable process that could support post penetration test hunts, intelligence driven hunts, vulnerability driven hunts, and leadership requested hunts. It was an attempt to bring structure to an area that did not yet have clear ownership or dedicated time.</p><p>In my current role, I often see threat hunts used reactively, usually after a compromise, when a business asks for a threat hunt across other systems to look for the same activity. That approach defeats the purpose of hunting. It can be useful during incident response to chase down trailheads, but the core principles of threat hunting are built on being proactive, not retrospective. Across many environments, the foundational pieces required for a strong hunting program are missing. Vulnerability management, asset management, asset ownership, clean detections, and a strong baseline all matter. They define the attack surface, clarify what is exposed, and establish who is accountable for it. Without those fundamentals, a threat hunt becomes guesswork instead of a disciplined process.</p><p>You can still threat hunt without all of those foundational pieces in place, but it becomes the least effective version of the discipline. Missing asset inventories, weak vulnerability management, unclear ownership, and noisy detections do not stop a hunt from happening. They stop it from being repeatable. They limit what you can prove, what you can trust, and what you can scale. You can still hunt, but you cannot build a reliable program on top of that.</p><p>When the foundational pieces are missing, it also multiplies the time spent on every hunt. Analysts end up collecting the same data repeatedly, checking the same systems, and rebuilding context that should already exist. Instead of progressing the hunt, most of the effort goes into compensating for gaps in visibility, ownership, or baseline understanding. The work becomes slower, less focused, and far more manual than it needs to be.</p><p>A structured hunt follows a clear workflow. A hypothesis is developed based on intelligence or observed patterns. Required data sources are identified. Searches or queries are designed to test the hypothesis. Results are analyzed to validate or disprove the idea. Pivot points are identified based on findings. Documentation captures the reasoning, evidence, and outcome. Each step is intentional and tied to the original hypothesis. None of it is random exploration.</p><p>All threat hunts and reports should be fully repeatable. Every data source used in the hunt needs to be collected, referenced, and documented so another analyst can reproduce the work without guessing. The findings from a hypothesis, whether positive, negative, or inconclusive, should be passed to the detection engineering team so the behavior can be monitored going forward. In some cases, even a no findings result can lead to a detection if the behavior has value or risk associated with it. And when a hunt cannot be converted into a detection, that should be documented as well, along with a schedule for when the hunt should be repeated based on business risk and relevance. This level of structure prevents analysts from chasing the same leads repeatedly and improves efficiency across the team. It also ensures that hunting contributes to long term visibility rather than becoming a one off exercise. Think of it like a penetration test without a report. The work might have been done, but without documentation, it is not useful to anyone.</p><p>A hunt is only valid if the environment has the telemetry to support it. Analysts must understand whether the required data actually exists, where it lives, how far back it goes, and what its limitations are. A hypothesis that cannot be tested with available data becomes a finding in itself. Threat hunting is not just about knowing what to look for. It is about knowing whether you even can look for it.</p><p>A structured hunting program consistently produces several operational benefits. These include improved detection coverage, validation of existing controls, identification of blind spots, increased visibility into the environment, strengthened incident response processes, and reduced exposure to threats. These benefits are not theoretical. They are outcomes I have seen across environments that apply structured, intelligence led hunting.</p><p>There are also common challenges that limit hunting effectiveness. These include limited time, limited data availability, limited intelligence alignment, limited expertise, false positives, and lack of dedicated resources. In environments with high alert volume, available time for structured hunts can be reduced. Hypothesis development can become difficult under pressure, and pivot decisions can become unclear without strong direction. These challenges highlight why structured processes and intelligence are essential.</p><p>Threat hunting is an analytical discipline. It requires slowing down to define the problem clearly, asking precise questions instead of broad ones, applying pivot discipline, avoiding unnecessary rabbit holes, documenting reasoning as part of the process, and using evidence to guide decisions. This mindset is what separates structured hunting from unstructured searching.</p><p>Threat hunting works because it forces an organization to confront what it can see, what it cannot, and what it has been assuming without evidence. When the process is structured, repeatable, and grounded in intelligence, it becomes one of the few disciplines that consistently exposes blind spots before an attacker does. That is the real value of threat hunting, and it is why the foundation matters more than the outcome of any single hunt.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=e6be7d19719a" width="1" height="1" alt="">]]></content:encoded>
        </item>
        <item>
            <title><![CDATA[My Practical SOC Analyst Professional (PSAP) Experience: A Practical Exam That Tested More Than My…]]></title>
            <link>https://medium.com/@brent.hachey/my-practical-soc-analyst-professional-psap-experience-a-practical-exam-that-tested-more-than-my-acb7a561a4da?source=rss-2a05f76d47f0------2</link>
            <guid isPermaLink="false">https://medium.com/p/acb7a561a4da</guid>
            <category><![CDATA[ai]]></category>
            <category><![CDATA[cybersecurity]]></category>
            <category><![CDATA[technology]]></category>
            <category><![CDATA[artificial-intelligence]]></category>
            <dc:creator><![CDATA[Brent Hachey]]></dc:creator>
            <pubDate>Thu, 04 Jun 2026 02:48:38 GMT</pubDate>
            <atom:updated>2026-06-04T02:48:38.823Z</atom:updated>
            <content:encoded><![CDATA[<p>My Practical SOC Analyst Professional (PSAP) Experience: A Practical Exam That Tested More Than My Skills</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/1024/1*OBEnRa2YM7UlVNYZiJT8MQ.jpeg" /></figure><p>Why I Chose the PSAP:</p><p>When I started looking at what to do next in my career, I knew I didn’t want another certification that was just multiple choice. I already had Security+ and CySA+, and they were good stepping stones, but they didn’t reflect the kind of work I was actually doing anymore. My employer had me sit in on a Google Mandiant practical threat hunting course, and that was the moment things shifted. Once you get hands on experience with real investigation work, it is hard to go back to memorizing terms and hoping the right answer is on the screen.</p><p>I wanted something practical. Something that felt like the job. Something that would actually make me better at what I do every day.</p><p>That is what pulled me toward the PSAP. I bought the certification at the end of 2025 because I wanted to push myself and grow as a security analyst. Then in January I landed a Threat Analyst II role on a leading MDR team, and the timing suddenly made sense. The PSAP was not just a goal anymore. It felt like the right next step to validate the work I was already doing.</p><p>Another thing that stood out to me was that the certification never expires. No renewals, no CEUs, no yearly fees. Just a practical exam that proves you can actually investigate an incident. And TCM Security has been growing fast in the hands on training space, so it felt like a credential that would only gain more weight over time.</p><p>I will admit something too. I had my eye on the early adopter badge. But imposter syndrome was loud. I kept thinking I was not ready or that I would get exposed somehow. It took a while to push past that feeling.</p><p>In the end, that was the reason I went for it. I wanted something real. Something challenging. Something that matched the work I was doing and forced me to grow. And I wanted to prove to myself that I could do it.</p><p>When Everything Went Wrong on Day One:</p><p>I sat down at 11 a.m. on a Saturday, full breakfast in me, feeling calmer than I ever did before a CompTIA exam. This one felt different. I was not sure what to expect, but I was not scared either. Maybe a little anxious to get started, mostly because I wanted that early adopters badge and I knew I had a free retake. It felt like I had room to breathe.</p><p>The setup hit me harder than I expected. It was raw. No detections. No alerts. No comforting SOC dashboards lighting up with obvious leads. This was a practical threat hunt exam, and it made that clear immediately.</p><p>I tried leaning on tools I had used in the SOC201 course, expecting them to guide me the same way they did in training. Instead, nothing worked the way I thought it would. Every attempt to streamline my workflow fell apart. Every tool I reached for felt wrong. My mind started playing tricks on me. The self doubt crept in fast, and before I knew it, I was spiraling into that familiar place where imposter syndrome waits with open arms.</p><p>I was so convinced something was broken that I reset the entire environment twice. Not once, but twice. I even emailed support, fully expecting them to confirm something was misconfigured. Instead, they told me everything was working as intended. That hit hard. It meant the problem was not the environment. It was me.</p><p>By that point, I had burned most of my first day. Hours gone. Confidence gone. I remember sitting there thinking I had made a mistake taking this exam. My mind told me I had already failed on day one.</p><p>By the time the clock hit 6 p.m., I was mentally exhausted. That was when something finally shifted. I had to stop and tell myself, stop doing what you think and start doing what you know. I do raw log analysis every day. I know how to hunt. I know how to pivot. I know how to follow the evidence.</p><p>The SOC201 course gives you options, but the exam forces you to choose. Once I made that mental shift, everything changed. By 8 p.m., I was already finding intrusive anomalies and unusual account activity across systems. The imposter in my head disappeared. I did not have time for him anymore. I was too busy building the attack chain.</p><p>And here is the part I wish someone had told me. You have to read the rules of engagement. Not just skim them. They shape your mentality. They tell you how to think, not what to find.</p><p>What the Exam Felt Like:</p><p>Once my mindset shifted on that first night and I finally found a rhythm that worked, the entire weekend took on a different energy. It was not easier, but it felt purposeful. I was not fighting myself anymore. I was working.</p><p>Day 2 started early at 10 a.m. and I went straight into refining my attack chain. I spent the whole day living inside system logs and the timeline editor, stitching together what happened and when. With every connection I made, my confidence grew, but I was still skeptical. I took screenshots of everything, documented every pivot, and wrote notes for every anomaly so I could refer back without losing my place. It felt like building a puzzle where the picture slowly comes into focus.</p><p>By the second night, I caught myself hunting for Easter eggs. I kept thinking, what did TCM hide, and that mindset pulled me deep into the weeds. I went through logs line by line, system by system, account by account, convinced there had to be something more. I fell down a rabbit hole of analysis and did not realize until the end of the night that I had spent way too much time chasing what might be there instead of focusing on what was actually next in my attack chain.</p><p>Day 3 was my final day in the environment. I finished the attack chain and then went through the entire thing again from start to finish. I wanted to make sure nothing was missing, nothing was assumed, and nothing was out of order. Once I was confident in the chain, I started my full timeline evidence collection and began drafting the report. I captured every IOC, every screenshot, every log location, and every step the threat actor took.</p><p>The reporting phase became its own marathon. I spent the first day of my reporting window finalizing everything. By the time I was done, I had written an eighty two page IR report that included the full attack chain, an IOC list, my methodology, my threat hunt hypothesis, and a detailed list of the tools I used. It was the most comprehensive report I had ever written, and it reflected every hour of effort that went into the investigation.</p><p>The Challenges I Ran Into:</p><p>The hardest part of the exam was not the technical work. It was keeping my mindset under control. Day One showed me how quickly doubt can take over. I spent hours convinced the environment was broken. I reset it twice and even emailed support. Losing that much time because I could not get out of my own head was a tough lesson.</p><p>Another challenge was over hunting. Once I found my rhythm, I pushed too far in the other direction. I started digging for hidden clues that were not there, going through logs line by line and chasing possibilities instead of following the actual attack chain. It felt productive in the moment, but it pulled me off track and cost me time I did not need to lose.</p><p>Imposter syndrome also showed up more than I expected. I do not have a university degree, and my background is a decade of IT roles across help desk, MSP work, and field service. During the exam, that old insecurity came back. I had to remind myself that I do this work every day and that I earned my place in this field.</p><p>The reporting phase brought its own challenge. Writing an eighty two page IR report after a full weekend of analysis was mentally draining. It forced me to slow down, stay organized, and explain every step clearly. It reminded me that investigation work does not end when the threat actor stops moving. It ends when the story is documented well enough for someone else to understand it.</p><p>These challenges shaped the experience more than anything else. They pushed me to stay disciplined, stay focused, and trust my process. The exam did not just test what I knew. It tested how I think under pressure.</p><p>How I Approached the Investigation:</p><p>During my study for the PSAP, one thing stood out right away. They allow open book. That changed how I prepared. Instead of trying to memorize every command or every pivot, I decided to build myself a personal runbook, the same way I do for work. I wanted something tailored to the exam and the course material, but also something that reflected how I actually investigate incidents in real life.</p><p>The purpose was not to hold my hand or replace what I did not know. It was to give me structure. I built out a threat hunt guide, IR playbook runs, and clear stop points to keep me from going too far down rabbit holes. I wanted a system that would keep me grounded when the pressure kicked in.</p><p>In the end, I did not use the runbook as much as I expected. But it was reassuring to know it was there. If I needed a quick PowerShell command or a Splunk query for something specific, I could grab it without breaking my flow. Just knowing I had it in my back pocket helped me stay calm.</p><p>I also built a quick threat hunt guide based on the Google Mandiant A4 framework. It was not a step by step checklist. It was more of a compass. Something to help guide me through the rocky parts of the exam and remind me to stay focused on the attacker’s behavior, not my assumptions.</p><p>My approach ended up being a mix of structure and instinct. The runbook gave me direction. The A4 framework gave me discipline. And the rest came down to trusting the skills I use every day as a threat analyst.</p><p>What Surprised Me:</p><p>Going into the exam, I knew it was not CTF based. I had done my research, I knew there was a rules of engagement document, and I knew there would be a threat actor report. Even with all of that, I still expected there to be some kind of starting point. Something small. A hint. A direction. Especially since the exam is tied to the SOC 201 course.</p><p>What surprised me most was how hands on it actually was. It ended up being more hands on than the Google Mandiant practical threat hunting course my employer put me through, and that course was already very real world. The PSAP leaned heavily into threat hunting, far more than I expected. Honestly, TCM undersells that part. The threat hunt focus is one of the strongest elements of the exam, and it caught me off guard in a good way. This is exactly the kind of exam I have been looking for.</p><p>Another thing that surprised me was the role they place you in. The preparation material says the exam is aimed at tier two and tier three analysts, but the scenario drops you directly into a corporate environment as a tier three security analyst. That shift in responsibility changes how you think. You are not just reacting. You are driving the investigation. You are expected to take ownership of the entire chain, not just pieces of it. I did not expect that level of autonomy, but it made the experience feel real.</p><p>The biggest surprise was how much the exam trusted me to think like an analyst instead of a student. There were no breadcrumbs. No nudges. No artificial hints. Just an environment, a threat actor, and the expectation that I could figure it out. That level of trust is rare in certification exams, and it made the whole experience feel more authentic than anything I have taken before.</p><p>The 82 Page Report:</p><p>TCM makes it clear from the start that the report is the heart of the exam. It is the only thing the grader sees. It is the only proof of what you found, how you found it, and how you think as an analyst. They highlight that the report shows your investigation, your methodology, and your ability to communicate clearly. That message stuck with me the entire time.</p><p>This was my first full IR report. I have written smaller threat hunt summaries before. I have done application reviews at my previous job. And in my current MDR role, I write a dozen case reports a day. But none of that prepared me for the scale of documenting an entire attack chain from start to finish.</p><p>My report ended up being eighty two pages because I wanted to include everything TCM required and then some. I wanted the grader to see every pivot, every IOC, every log reference, every screenshot, and every conclusion backed by evidence. I did not want anything to be assumed or left to interpretation. If I found it, I documented it.</p><p>Writing the report forced me to slow down and think through the investigation again. It made me revisit my notes, my screenshots, and my timeline. It made me justify why I believed each step happened the way it did. It made me explain the attacker’s behavior in a way that someone else could follow without ever touching the environment.</p><p>The report became more than a requirement. It became the part of the exam where everything clicked. It showed me how much I had actually learned. It showed me that I could take a complex investigation and turn it into a clear narrative. It showed me that I was capable of writing a real IR report, not just an exam submission.</p><p>By the time I finished, the length did not matter. What mattered was that it reflected the work I put in and the analyst I am becoming.</p><p>The Waiting Period:</p><p>The waiting period was rough. I am used to getting exam results right away, so having to sit and wait felt completely foreign. TCM gives a grading window, and support tells you it can take five to seven business days. When you have just poured an entire weekend into an investigation and an eighty two page report, that window feels like an eternity.</p><p>This is where I felt conflicted. On one hand, the wait was stressful. On the other hand, the reason for the delay is that a real human reads your report. Someone actually goes through your evidence, your screenshots, your attack chain, and your methodology. That adds a huge level of respect to the process. If they tried to guarantee fast turnaround times with hard deadlines, it would almost certainly push them toward automated grading. And once AI gets involved in scoring, people would start questioning the quality and fairness of the results. That is just my opinion, but I am glad a human marked my exam.</p><p>I ended up waiting three business days before I got my pass email. It arrived in the evening, which might have been due to time zone differences. When I opened it, I noticed there was no score included. I reached out to support and they told me:</p><p>“Please note that the TCM’s grading system only publishes pass and fail as results.</p><p>To maintain fairness to all students, we are unable to provide any detailed evaluation beyond this.”</p><p>I will be honest. After working that hard, I was looking forward to seeing a final grade. Even just a number. But at the end of the day, I am happy with the certification and the experience. The waiting period definitely varies, and it is the one part of the process that could use some smoothing out. Still, knowing a real person reviewed my work made the pass feel more meaningful.</p><figure><img alt="" src="https://cdn-images-1.medium.com/max/840/1*eyTVLfefTkT0h0dfv67qAQ.jpeg" /></figure><h3>Would I Recommend It:</h3><p>I am not going to decide who this exam is for. That is up to the student and TCM. There is already plenty of information out there about the intended audience. What I can speak to is whether I would personally recommend it based on my own experience.</p><p>TCM certifications are still lesser known in the HR world. Anyone taking their exams already understands that. It was one of the reasons I wanted the early adopters badge and why I wanted to share my experience publicly. I wanted other analysts and people trying to break into the field to see what this exam actually feels like, not just what the website says.</p><p>So here is my honest take.</p><p>If you are doing this exam strictly for résumé keywords or to get past automated filters, this is probably not the certification that will move the needle on its own. But if you are doing it for the skill growth, for the challenge, for the ability to walk someone through a full threat hunt and an IR report from start to finish, then yes, I would recommend it.</p><p>This exam makes you think. It forces you to investigate, document, and communicate like an actual analyst. It gives you hands on experience that you can talk about in an interview or on the job. And I genuinely think TCM is gaining traction. They are becoming a leader in practical, hands on learning, and I can see them eventually surpassing Security Blue Team in that space.</p><p>If your goal is to grow and to prove something to yourself, this exam delivers.</p><p>What I Plan To Do Next:</p><p>I am one of many people who believe certifications only move the needle so far. They help, but they are not the whole picture. I am trying to balance skill growth with proven certifications while actually putting those skills to use. I also do not plan to stack certifications from one vendor. I want a mix of perspectives, not a wall of the same logo.</p><p>My employer pays for BTLO for senior analysts, and that is something I want to work toward. In the meantime, I plan to work through BTLO labs on my own and focus more deeply on DFIR. I still have a few TCM courses that I can use to build onto my current skill set, and I want to take advantage of those when the time is right.</p><p>I also have a personal interest in developing deeper OSINT skills. I think it would round out my abilities and give me another angle when approaching investigations.</p><p>To be honest, I am taking a short break from courses right now. I want to focus on my personal development and my growth as a Threat Analyst II in my new role. There is a lot to learn on the job, and I want to give myself the space to absorb it.</p><p>Closing Statement:</p><p>Just remember that we all grow at different speeds and we are all on our own path. The only person you should compare yourself to is your past self. We all come from different walks of life, and only the universe knows what comes next from the choices we make.</p><p>Imposter syndrome is not a sign that you do not belong. It is self reflection mixed with fear. It is a tool you can use to push yourself toward becoming the best version of who you can be.</p><p>What This Certification Meant to Me:</p><p>I am proud to share that I earned the Practical SOC Analyst Professional certification. The experience pushed me in ways I did not anticipate and reminded me why I enjoy this field. Finishing as part of the early group taking on this new certification made the whole thing feel even more meaningful.</p><p>Now that the dust has settled, I am looking forward to catching my breath and continuing to grow in incident response and threat hunting.</p><p>Thank you to TCM Security and Andrew Prince for so much knowledge.</p><img src="https://medium.com/_/stat?event=post.clientViewed&referrerSource=full_rss&postId=acb7a561a4da" width="1" height="1" alt="">]]></content:encoded>
        </item>
    </channel>
</rss>