<?xml version="1.0" encoding="utf-8"?><feed xmlns="http://www.w3.org/2005/Atom" xml:lang="en"><generator uri="https://jekyllrb.com/" version="3.10.0">Jekyll</generator><link href="https://securityvolition.com/blog/feed.xml" rel="self" type="application/atom+xml" /><link href="https://securityvolition.com/" rel="alternate" type="text/html" hreflang="en" /><updated>2026-11-15T12:34:01-05:00</updated><id>https://securityvolition.com/blog/feed.xml</id><title type="html">Security Volition Blog</title><subtitle>Practitioner notes on incident response, threat trends, and the defensive controls that actually reduce damage.</subtitle><author><name>Richard Brackett</name></author><entry><title type="html">What happens when the SaaS Platform Running Your Business Goes Dark?</title><link href="https://securityvolition.com/blog/2026/11/15/what-happens-when-the-saas-platform-running-your-business/" rel="alternate" type="text/html" title="What happens when the SaaS Platform Running Your Business Goes Dark?" /><published>2026-11-15T12:15:06-05:00</published><updated>2026-11-15T12:15:06-05:00</updated><id>https://securityvolition.com/blog/2026/11/15/what-happens-when-the-saas-platform-running-your-business</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/15/what-happens-when-the-saas-platform-running-your-business/"><![CDATA[<h2 id="sources">Sources</h2>

<ul>
  <li>Microsoft 365 outage (AP): https://apnews.com/article/microsoft-outage-outlook-email-365-recover-ca395c0f1e51af784b6d70afb6554f6a</li>
  <li>Microsoft 365 outage (CNBC): https://www.cnbc.com/2026/11/12/microsoft-outage-outlook-aws-google.html</li>
  <li>Workday outage: https://www.webpronews.com/workday-outage-disaster-recovery-test-production-data/</li>
  <li>Salesforce outage (CIO): https://www.cio.com/article/4223037/salesforces-massive-outage-exposes-the-hidden-risks-of-cloud-dependencies.html</li>
  <li>UMMC Epic ransomware (Healthcare Dive): https://www.healthcaredive.com/news/university-of-mississippi-medical-center-ransomware-attack/812823/</li>
  <li>UMMC Epic ransomware (BleepingComputer): https://www.bleepingcomputer.com/news/security/mississippi-medical-center-reopens-clinics-hit-by-ransomware-attack/</li>
  <li>Change Healthcare (AHA survey): https://www.aha.org/2024-03-15-aha-survey-change-healthcare-cyberattack-significantly-disrupts-patient-care-hospitals-finances</li>
  <li>CDK Global (Reuters): https://www.reuters.com/technology/cybersecurity/ripple-effect-cdk-hack-widens-more-us-auto-dealers-flag-hit-2024-06-24/</li>
</ul>

<h2 id="related">Related</h2>

<ul>
  <li>[[2026-11-10 LinkedIn Post 2 - When Your RMM Vendor Goes Dark]]</li>
  <li>[[2026-11-10 LinkedIn Post 1 - Keep the Admin UI Off the Internet]]</li>
  <li>[[2026-11-09 LinkedIn Article - AI-Speed Exploitation and Compensating Controls]]</li>
</ul>]]></content><author><name>Richard Brackett</name></author><category term="article" /><category term="business-continuity" /><category term="saas-outage" /><category term="third-party-risk" /><category term="microsoft-365" /><category term="workday" /><category term="salesforce" /><category term="ehr" /><summary type="html"><![CDATA[Sources]]></summary></entry><entry><title type="html">Agent Skills Are the New Browser Extensions</title><link href="https://securityvolition.com/blog/2026/11/15/agent-skills-are-the-new-browser-extensions/" rel="alternate" type="text/html" title="Agent Skills Are the New Browser Extensions" /><published>2026-11-15T08:41:01-05:00</published><updated>2026-11-15T08:41:01-05:00</updated><id>https://securityvolition.com/blog/2026/11/15/agent-skills-are-the-new-browser-extensions</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/15/agent-skills-are-the-new-browser-extensions/"><![CDATA[<p>Acronis researchers found more than 575 malicious OpenClaw skills posing as ordinary utilities. They delivered trojans, cryptominers, and infostealers to Windows and macOS machines.</p>

<p>The uncomfortable part: an agent skill is instructions plus executable code, and it runs with whatever permissions the agent already has. Install a malicious skill and you've handed an attacker your agent's access to files, tokens, and connected SaaS applications. The same report describes an operator whose AI tooling was configured to run without approval prompts. Plenty of defenders do that too, for convenience.</p>

<p>For responders, this is the new browser-extension problem, and it needs the same answer: an allowlist for agent skills and plugins, backed by least-privilege agent identities.</p>

<p>Done right, that's:</p>
<ul>
  <li>Skills come from an approved internal catalog, pinned to reviewed versions, not pulled ad hoc from public marketplaces</li>
  <li>Each agent runs under its own scoped identity with short-lived credentials, not a developer's full session</li>
  <li>Human approval stays on for anything that executes code, touches credentials, or sends data out</li>
  <li>Skill installs and agent tool calls are logged where your IR team can actually query them</li>
</ul>

<p>When an agent workstation gets compromised, the first scoping question is simple: what could this agent reach?</p>

<p>https://www.helpnetsecurity.com/2026/11/10/acronis-ai-agent-cyberattacks-report/</p>

<p>#AISecurity #DFIR #IncidentResponse</p>]]></content><author><name>Richard Brackett</name></author><category term="ai-security" /><category term="agent-skills" /><category term="least-privilege" /><summary type="html"><![CDATA[Acronis researchers found more than 575 malicious OpenClaw skills posing as ordinary utilities. They delivered trojans, cryptominers, and infostealers to Windows and macOS machines.]]></summary></entry><entry><title type="html">When Your RMM Vendor Goes Dark</title><link href="https://securityvolition.com/blog/2026/11/13/when-your-rmm-vendor-goes-dark/" rel="alternate" type="text/html" title="When Your RMM Vendor Goes Dark" /><published>2026-11-13T14:54:01-05:00</published><updated>2026-11-13T14:54:01-05:00</updated><id>https://securityvolition.com/blog/2026/11/13/when-your-rmm-vendor-goes-dark</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/13/when-your-rmm-vendor-goes-dark/"><![CDATA[<p>N-able took its hosted N-central environment offline on Nov 9 after a researcher published 33 critical vulnerabilities, chainable into four exploits dubbed Able2Shell. On-prem customers were told to take their servers offline too. N-able said it hadn't seen exploitation.</p>

<p>That was the right call, and a really hard one. Now picture it from the MSP side: the tool you use to patch, script, and remote into every client goes dark with no firm ETA, and your clients are asking whether they're exposed.</p>

<p>This is where IR delivery earns trust or loses it. The control is a tested critical-vendor outage runbook for your management plane. What I'd want to see:</p>
<ul>
  <li>A documented, out-of-band way to reach and patch managed endpoints when the RMM is down</li>
  <li>Pre-written client notices that say what you know, what you don't, and when you'll update next</li>
  <li>A scoping checklist ready before you power back on: logins since the disclosure, new admin accounts, scripts pushed, unexpected remote sessions</li>
</ul>

<p>That last item matters. In earlier N-central exploitation, attackers pushed scripts and tools downstream into managed environments. One console, many victims.</p>

<p>And it happened again this week. On Nov 11, per Huntress, attackers abused a third-party ScreenConnect extension called RMM+, which accepted a blank access key and skipped login and MFA, to push a toolkit called SCKit to at least 288 endpoints across 48 orgs in about 90 minutes. Your RMM's plugins and extensions belong in that runbook too. Who on your team is tracking those?</p>

<p>Tabletop the day your RMM vendor goes dark. It just happened.</p>

<p>https://www.huntress.com/blog/n-able-disables-cloud-services-following-disclosure-of-able2shell-exploit-chains</p>

<p>#IncidentResponse #MSP #ThirdPartyRisk</p>]]></content><author><name>Richard Brackett</name></author><category term="n-able" /><category term="rmm" /><category term="third-party-risk" /><category term="customer-comms" /><summary type="html"><![CDATA[N-able took its hosted N-central environment offline on Nov 9 after a researcher published 33 critical vulnerabilities, chainable into four exploits dubbed Able2Shell. On-prem customers were told to take their servers offline too. N-able said it hadn't seen exploitation.]]></summary></entry><entry><title type="html">Supermarket Freezer Login Pages</title><link href="https://securityvolition.com/blog/2026/11/12/supermarket-freezer-login-pages/" rel="alternate" type="text/html" title="Supermarket Freezer Login Pages" /><published>2026-11-12T08:17:01-05:00</published><updated>2026-11-12T08:17:01-05:00</updated><id>https://securityvolition.com/blog/2026/11/12/supermarket-freezer-login-pages</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/12/supermarket-freezer-login-pages/"><![CDATA[<p>About 700 French supermarkets lost their freezers to an attacker last month, and nobody needed a zero-day.</p>

<p>Luke de Wolf's write-up of the refrigeration attacks is worth your time. Starting the night of Oct 11-12, attackers changed settings on internet-exposed refrigeration controllers. CERT-FR named Danfoss, Copeland and Dixell. A second wave hit other European countries overnight into Nov 2. Attribution is still unknown.</p>

<p>ANSSI's director told the French Senate the way in was internet exposure, unprotected admin access and weak or default passwords, not the published CVEs.</p>

<p>Why were they online? Mostly so the refrigeration contractor could monitor and manage them remotely. That's the part every IR team should take home. Your OT exposure is often your vendor's remote access. (shakes head and sighs)</p>

<p>Two data points stand out:</p>
<ul>
  <li>Censys counted 1,825 exposed Danfoss StoreView interfaces in France before the attack and 220 by Nov 2. Operators can fix this fast once they have a reason to.</li>
  <li>In Estonia, Selver's refrigeration partner responded within 10 minutes. Temperatures rose only 2-3 degrees and no product was lost. Monitoring plus a partner who answers works.</li>
</ul>

<p>These controllers aren't only a European problem. Censys says the US has the third-most exposed StoreView interfaces, behind France and Belgium.</p>

<p>What I'd do this week:</p>
<ol>
  <li>Inventory building and OT management interfaces reachable from the internet. Check HVAC, refrigeration and access control, not just servers.</li>
  <li>For each one, name the contractor who uses it and how. Move that access behind a VPN or protected network. Change default passwords.</li>
  <li>Put vendor response times in the contract and test them. Selver's 10 minutes didn't happen by accident.</li>
  <li>Add OT to your IR playbooks. Who can pull a controller offline at 3 AM? Does cutting the contractor's data feed affect your insurance? CERT-FR says to check before you cut it.</li>
  <li>After an incident, review the whole configuration, including defrost schedules. Replace end-of-life units. A factory reset alone leaves them vulnerable.</li>
</ol>

<p>If an attacker can reach a login page, assume they will try it.</p>

<p>https://x.com/lukedewolf/status/2117635330710941996</p>

<p>#DFIR #IncidentResponse #OTSecurity</p>]]></content><author><name>Richard Brackett</name></author><category term="dfir" /><category term="ot-security" /><category term="third-party-risk" /><category term="incident-response" /><summary type="html"><![CDATA[About 700 French supermarkets lost their freezers to an attacker last month, and nobody needed a zero-day.]]></summary></entry><entry><title type="html">Keep the Admin UI Off the Internet</title><link href="https://securityvolition.com/blog/2026/11/11/keep-the-admin-ui-off-the-internet/" rel="alternate" type="text/html" title="Keep the Admin UI Off the Internet" /><published>2026-11-11T11:38:01-05:00</published><updated>2026-11-11T11:38:01-05:00</updated><id>https://securityvolition.com/blog/2026/11/11/keep-the-admin-ui-off-the-internet</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/11/keep-the-admin-ui-off-the-internet/"><![CDATA[<p>Barracuda confirmed recently that two Email Security Gateway flaws, CVE-2026-116606 and CVE-2026-111560, are being exploited against a small number of organizations. Chained, they give an attacker root on the appliance.</p>

<p>The important point, though, is exploitation <em>requires</em> access to the administrative interface. That's a precondition you, the defender, control.</p>

<p>Patch, yes. ESG 9.4.0.030 or later, and turn automatic patching back on if someone turned it off. But the control that makes this nearly irrelevant is management-plane isolation: admin interfaces for email gateways, firewalls, VPNs, and load balancers are reachable only from a dedicated management network or jump host, never from the internet. Here's an idea: how about a jump box in Azure, AWS, or other IaaS provider with a fixed IP address that you can allow-list on the appliances. And not just the appliances, how about an inbound ACL on the Internet link that drops anything to the management interface that isn't your fixed IP? Leave the jump box offline until you need it and furthermore, secure it with MFA-enabled access to get to it.</p>

<p>What this should look like:</p>
<ul>
  <li>A scheduled external scan that alerts when any admin UI answers from outside, because change management isn't 100%, amiright?</li>
  <li>Admin access through a hardened jump host with phishing-resistant MFA</li>
  <li>Admin logins and config changes forwarded to your SIEM, not just stored on the box. Miss this one at your peril.</li>
</ul>

<p>If your ESG admin UI was exposed before you patched, treat it as a threat hunting exercise, not a closed ticket. Review config changes, restored backups, and new accounts since early November.</p>

<p>Edge appliances keep getting hit because their management planes sit on the internet. That part is fixable.</p>

<p>https://trust.barracuda.com/security/pre-authentication-authorization-bypass-command-injection</p>

<p>#DFIR #IncidentResponse #EdgeSecurity</p>]]></content><author><name>Richard Brackett</name></author><category term="barracuda" /><category term="edge-devices" /><category term="management-plane" /><summary type="html"><![CDATA[Barracuda confirmed recently that two Email Security Gateway flaws, CVE-2026-116606 and CVE-2026-111560, are being exploited against a small number of organizations. Chained, they give an attacker root on the appliance.]]></summary></entry><entry><title type="html">Fast Inference AI Intrusions</title><link href="https://securityvolition.com/blog/2026/11/10/fast-inference-ai-intrusions/" rel="alternate" type="text/html" title="Fast Inference AI Intrusions" /><published>2026-11-10T10:23:01-05:00</published><updated>2026-11-10T10:23:01-05:00</updated><id>https://securityvolition.com/blog/2026/11/10/fast-inference-ai-intrusions</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/10/fast-inference-ai-intrusions/"><![CDATA[<p>If your containment step needs a human to approve it, an AI-run intrusion will probably be finished before anyone does.</p>

<p>Ben at MOXFIVE published a sharp piece on this. A few points from it that every IR lead should read:</p>

<ul>
  <li>Machine-speed attacker activity went from about one in six of their investigations in Q4 2025 to more than six in ten by Q3 2026.</li>
  <li>Attackers are now running many fronts at once: pulling data, minting credentials, creating accounts and changing permissions in the same minutes.</li>
  <li>Fast inference (1,000+ tokens per second) is already for sale. When they re-timed real cloud intrusions at that speed, one run from a hijacked developer account to production secrets dropped from about 11 minutes to as little as 2.</li>
  <li>Cloud audit logs often take around five minutes just to arrive. Five to fifteen minutes from action to alert is common.</li>
</ul>

<p>Do the math: the first alert can describe something that already happened.</p>

<p>What I'd do now, before that day arrives:</p>

<ol>
  <li>Pre-authorize containment. Decide in writing which actions run without sign-off once an attack is confirmed: disable the identity, revoke sessions and tokens, isolate the endpoint, lock down the cloud role. Get legal and the business to agree now, not at 2 AM.</li>
  <li>Measure your pipeline, not just your analysts. Time action to log, log to alert, alert to containment. Most teams only track the last one.</li>
  <li>Prove you can contain in the cloud control plane, not only on endpoints. Test it.</li>
  <li>Shrink what's exposed. A faster attacker gets more from every reachable system. Harel's advice is to put an identity-aware edge in front of public apps.</li>
  <li>Plan scoping for parallel workstreams. Thousands of interleaved requests from many addresses won't untangle by hand.</li>
</ol>

<p>The decisions that matter have to be made before the incident, not during it.</p>

<p>https://x.com/benharelone/status/2117651795610939555</p>

<p>#DFIR #IncidentResponse #AISecurity</p>]]></content><author><name>Richard Brackett</name></author><category term="dfir" /><category term="ai-security" /><category term="incident-response" /><category term="containment" /><summary type="html"><![CDATA[If your containment step needs a human to approve it, an AI-run intrusion will probably be finished before anyone does.]]></summary></entry><entry><title type="html">One Place for DFIR and Blue-Team CTFs</title><link href="https://securityvolition.com/blog/2026/11/09/one-place-for-dfir-and-blue-team-ctfs/" rel="alternate" type="text/html" title="One Place for DFIR and Blue-Team CTFs" /><published>2026-11-09T14:09:30-05:00</published><updated>2026-11-09T14:09:30-05:00</updated><id>https://securityvolition.com/blog/2026/11/09/one-place-for-dfir-and-blue-team-ctfs</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/09/one-place-for-dfir-and-blue-team-ctfs/"><![CDATA[<p>Most CTF calendars are built for red teamers. If you work the defensive side, finding forensics and blue-team practice usually means digging through general listings to see which ones include forensics challenges.</p>

<p>So I built one place for it: the Security Volition CTF Board, https://ctf.securityvolition.com</p>

<p>It tracks upcoming DFIR, forensics, and blue-team CTFs with confirmed dates. The next few weeks include WolvCTF, SANS DFIR NetWars in Paris and Amsterdam, and the Guardians blue-team qualifier in February. Events whose dates aren't confirmed yet, like the SANS Holiday Hack Challenge and the Magnet Virtual Summit CTF, sit in a separate section. It also lists always-on labs like CyberDefenders, Blue Team Labs Online, and Hack The Box Sherlocks for when there's no event on the calendar.</p>

<p>The defensive angle: what IR teams are seeing is that the gap usually isn't tooling, it's reps. Analysts who regularly work timelines, memory, and log evidence under a clock move faster when it's real. A few hours of forensics CTF a month is cheap practice for that.</p>

<p>If I'm missing an event, tell me and I'll add it.</p>

<p>#DFIR #IncidentResponse #DigitalForensics #CTF #BlueTeam</p>]]></content><author><name>Richard Brackett</name></author><category term="dfir" /><category term="ctf" /><category term="blue-team" /><category term="training" /><summary type="html"><![CDATA[Most CTF calendars are built for red teamers. If you work the defensive side, finding forensics and blue-team practice usually means digging through general listings to see which ones include forensics challenges.]]></summary></entry><entry><title type="html">The Help Desk Call Is the Breach</title><link href="https://securityvolition.com/blog/2026/11/09/the-help-desk-call-is-the-breach/" rel="alternate" type="text/html" title="The Help Desk Call Is the Breach" /><published>2026-11-09T13:10:00-05:00</published><updated>2026-11-09T13:10:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/09/the-help-desk-call-is-the-breach</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/09/the-help-desk-call-is-the-breach/"><![CDATA[<p>Cisco Talos's Q3 2026 IR trends report has a finding every service desk leader should read: attackers are calling users while impersonating the IT help desk, then talking them into opening a remote session with a legitimate tool like Microsoft Quick Assist. In one engagement, that call was the start of a ransomware intrusion. We also see calls like this regularly leading to BEC's. We have encountered one specialized group that focuses on hotel property management systems as well. They call the front desk at night and convince the night shift to allow remote access. Then they move on to fraudulent charge-backs to a credit card the attacker controls. It ain't millions, but there are a LOT of hotels.</p>

<p>Phishing still led initial access, at 37% of Talos engagements, and ransomware and pre-ransomware rose to 27%. What stands out is that the weak point wasn't a missing tool. It was a process that let a voice on the phone stand in for identity.</p>

<p>Identity must be verified in both directions. Users need a known way to confirm a call really is IT, such as a callback through the published number or a ticket they can see. The help desk needs out-of-band verification before any MFA reset, re-enrollment, or remote session, with extra checks for privileged accounts. Have a scripted verification step nobody skips, and remote-access tools that are allow-listed so a new one launching raises an alert. You would see that alert, right?</p>

<p>In post-incident briefings, the fix that sticks is usually a process change, not a new product.</p>

<p>#IncidentResponse #SocialEngineering #Ransomware</p>

<p>https://blog.talosintelligence.com/ir-trends-q3-2026/</p>]]></content><author><name>Richard Brackett</name></author><category term="leadership" /><category term="customer-facing" /><category term="social-engineering" /><category term="ransomware" /><summary type="html"><![CDATA[Cisco Talos's Q3 2026 IR trends report has a finding every service desk leader should read: attackers are calling users while impersonating the IT help desk, then talking them into opening a remote session with a legitimate tool like Microsoft Quick Assist. In one engagement, that call was the start of a ransomware intrusion. We also see calls like this regularly leading to BEC's. We have encountered one specialized group that focuses on hotel property management systems as well. They call the front desk at night and convince the night shift to allow remote access. Then they move on to fraudulent charge-backs to a credit card the attacker controls. It ain't millions, but there are a LOT of hotels.]]></summary></entry><entry><title type="html">Know What You Run. Patch Faster. Plan for the Day Your Security Control Fails.</title><link href="https://securityvolition.com/blog/2026/11/09/know-what-you-run-patch-faster/" rel="alternate" type="text/html" title="Know What You Run. Patch Faster. Plan for the Day Your Security Control Fails." /><published>2026-11-09T08:32:00-05:00</published><updated>2026-11-09T08:32:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/09/know-what-you-run-patch-faster</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/09/know-what-you-run-patch-faster/"><![CDATA[<p>OpenAI's new GPT-6 Cyber model matters less as an AI headline than as a signal about how fast vulnerability discovery is moving — and what that means if you own a network.</p>

<p>According to SecurityWeek's reporting on OpenAI's announcement, GPT-6 Cyber scored 48.5% on ExploitGym, up from 32.1% for GPT-5.6 Cyber. Pointed at the Linux kernel with the same tooling OpenAI used for its predecessor, it reproduced 1,724 vulnerabilities and built functional privilege-escalation exploits for 948 of them. GPT-5.6 Sol produced 166. In Firefox, it documented 81 new flaws that Mozilla rated high severity, each with evidence and reproducible tests. OpenAI says it has reported 304 critical issues to maintainers with proposed fixes. Access is limited for now to select approved Daybreak Red customers.</p>

<p>OpenAI also stresses alignment and scope discipline, and Dragos helped pressure-test safeguards around industrial systems. That governance work matters. It does not change the operational takeaway: the skill floor for reconnaissance and vulnerability research is dropping. Rob T. Lee, Dragos's chief AI officer, said as much — and that point belongs on every vulnerability-management backlog review this quarter.</p>

<p>If AI-assisted research can compress the path from interesting bug to working exploit, three old disciplines stop being optional. Know what you run. Get patches on faster. And assume that critical devices — especially the security controls themselves — will eventually fail under rapid discovery and exploitation.</p>

<p>Really understand your software (and hardware) inventory, which is a huge challenge in almost any environment, but it is becoming more and more vital.</p>

<p>You cannot patch what you do not know you run. That sounds basic until you inventory what actually sits on the network: forgotten appliance web UIs, temporary jump boxes that became permanent, container images still shipping musl and OpenSSH, switches with management reachable from the wrong VLAN, VPN gateways, email gateways, EDR consoles, identity connectors, and firmware unchecked since the last refresh.</p>

<p>For this threat model, inventory means:</p>

<ul>
  <li>A living list of internet-facing and edge systems, with product, version, and who can authorize an emergency change.</li>
  <li>Firmware and appliance software treated as first-class assets, not "network team stuff."</li>
  <li>Non-human and service identities those products use, so you know what to revoke when a box is suspect.</li>
  <li>A clear answer to what terminates VPN, what is your IdP, what is your EDR management plane, and where OpenSSH still listens — without a week of meetings.</li>
</ul>

<p>If a high-severity finding drops on a product family tomorrow, the organizations that respond in hours already know where that product lives.</p>

<p>Be highly reactive to new security patches and understand it may already be too late</p>

<p>The discovery-to-exploit window is shrinking, as it has been for years now; this is just a new acceleration. This past week made the point without a lab demo.</p>

<p>CVE-2026-19868 in SonicWall SonicOS SSL-VPN can let someone with valid credentials for an alternate auth method bypass SAML-enforced MFA when SAML is enabled. SonicWall investigated reports of exploitation. CISA added it to the Known Exploited Vulnerabilities catalog. There is a fix; there is no workaround. The practical hunt is VPN logins that never produced a matching IdP sign-in and MFA event.</p>

<p>CVE-2026-123158 — "SSHole" in OpenSSH — can allow login with any password between 257 and 1,024 characters on certain musl-based systems without PAM, such as Alpine, OpenWrt, and Yocto. Accounts that should never accept a password, including root in some configurations, become reachable. Patches went out; within days WatchTowr, Previdian, GreyNoise, and others reported exploitation attempts, initially tied to botnets and miners.</p>

<p>On Zyxel GS1900 switches, researchers showed that firmware patched for an earlier flaw (CVE-2026-7273) still fell to an exploit path now tracked as CVE-2026-107441. GreyNoise had already reported data theft from nearly 1,000 switches. Patching once was not enough; the follow-on fix had to land too.</p>

<p>That pattern is the new normal: disclose or detect, weaponize quickly, hit edge and embedded gear first. Patch SLAs built for quarterly change windows do not match that tempo — especially for internet-facing services, VPN and firewall appliances, hypervisors, identity infrastructure, and anything that brokers trust into the LAN.</p>

<p>Reactive does not mean reckless. It means pre-approved emergency change paths, tested rollback, and a triage rule that internet-facing and security-control flaws jump the queue the same day credible exploitation appears.</p>

<p>Revisit compensating controls — especially when the broken thing is a security control.</p>

<p>Patching is necessary. It is not sufficient. Architecture decides how much an attacker gets when the next bypass lands, or when the appliance that was supposed to protect you is the one that got owned.</p>

<p>Security products are high-value targets because they are trusted. VPNs, firewalls, email gateways, EDR consoles, and identity providers sit on critical paths. When one fails, you need controls that do not live inside that box.</p>

<p>Worth funding this quarter:</p>

<ul>
  <li>Treat decrypted VPN and remote-access traffic as untrusted. Enforce policy at a boundary the VPN gateway does not control — segmentation, an independent downstream firewall, or an allowlist of destinations and ports. VLANs on the same compromised appliance are not independence.</li>
  <li>Pull admin interfaces off the internet. Management planes for firewalls, switches, hypervisors, and security tools should require a jump path, not a public URL.</li>
  <li>Prefer phishing-resistant MFA at the identity provider. Do not treat MFA on the appliance alone as the whole story. "MFA enabled" and "MFA actually enforced on every successful login" are different statements.</li>
  <li>Send logs somewhere the appliance cannot erase — VPN, IdP, firewall, and EDR admin logs off-box.</li>
  <li>Least privilege everywhere the security tool can reach: no broad LAN from the VPN pool, no standing domain admin on jump hosts, no oversized service accounts.</li>
  <li>Defense in depth so one failed control is not the whole game. If the email gateway is bypassed, attachment detonation and identity signals still matter. If the EDR console is suspect, you still need network telemetry and a way to isolate hosts without trusting that console alone.</li>
  <li>A tested IR and containment plan for "our security tool is the compromised thing." Who pulls the VPN offline. Who rotates secrets the appliance could have seen. How you hunt with logs that do not depend on the box. Practice that tabletop before you need it at 2 a.m.</li>
</ul>

<p>None of those controls stop every bypass. They shrink blast radius and buy detection time.</p>

<p>What to do this quarter</p>

<ol>
  <li>Produce a one-page inventory of edge, internet-facing, and security-control systems: product, version, owner, emergency patch path.</li>
  <li>Rewrite patch SLAs for that list. Aim for hours-to-days when a flaw is exploited or hits KEV — not "next maintenance window."</li>
  <li>For each VPN, firewall, email gateway, EDR, and IdP: name one compensating control that still works if that product is compromised, and verify it is deployed.</li>
  <li>Confirm admin UIs are not internet-exposed; fix any that are.</li>
  <li>Verify off-box logging for those same products, and that someone is alerted on auth anomalies (for example, VPN success without IdP MFA).</li>
  <li>Run one tabletop: edge security appliance exploited; assume credentials and config exposed. Capture gaps in containment and recovery.</li>
  <li>Watch AI-assisted vuln research output the way you already watch CISA KEV — as an early signal to re-check inventory, not as science fiction.</li>
</ol>

<p>AI will keep raising how fast flaws can be found and turned into working exploits. Vendors will keep shipping patches. Your job is the boring middle: know what you run, move when the patch drops, and design so the day a critical control fails, it is an incident — not an outage of your whole defensive strategy.</p>

<p>That work is not glamorous. It is what still works when the exploit window is measured in days, not months.</p>

<p>https://www.securityweek.com/openai-says-new-gpt-6-cyber-model-hunts-flaws-without-crossing-the-line/<br />
https://www.helpnetsecurity.com/2026/11/09/exploited-sonicwall-ssl-vpn-flaw-lets-attackers-bypass-saml-enforced-mfa-cve-2026-19868/<br />
https://www.securityweek.com/hackers-exploiting-sshole-authentication-bypass-in-openssh/<br />
https://thehackernews.com/2026/11/zyxel-gs1900-patch-fails-to-block-zero.html<br />
https://www.zyxel.com/global/en/support/security-advisories/zyxel-security-advisory-for-out-of-bounds-write-vulnerability-in-gs1900-series-switches-11-03-2026</p>

<p>#DFIR #IncidentResponse #VulnerabilityManagement #CompensatingControls #CyberSecurity</p>]]></content><author><name>Richard Brackett</name></author><category term="article" /><category term="vulnerability-management" /><category term="compensating-controls" /><category term="inventory" /><category term="patching" /><category term="gpt-6-cyber" /><category term="edge-devices" /><summary type="html"><![CDATA[OpenAI's new GPT-6 Cyber model matters less as an AI headline than as a signal about how fast vulnerability discovery is moving — and what that means if you own a network.]]></summary></entry><entry><title type="html">Low-Severity ESXi Bug, High-Severity Lesson</title><link href="https://securityvolition.com/blog/2026/11/07/low-severity-esxi-bug-high-severity-lesson/" rel="alternate" type="text/html" title="Low-Severity ESXi Bug, High-Severity Lesson" /><published>2026-11-07T09:40:00-05:00</published><updated>2026-11-07T09:40:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/07/low-severity-esxi-bug-high-severity-lesson</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/07/low-severity-esxi-bug-high-severity-lesson/"><![CDATA[<p>A "low-severity" CVE just reminded me that severity scores are not threat models.</p>

<p>CVE-2026-41709 is an ESXi insufficient-logging flaw: an attacker with admin access can perform certain operations without being logged. CISA added it to KEV last week, and CrowdStrike reported that the China-nexus actor Warp Panda used it as a zero-day to reduce audit-trail visibility in intrusions at US organizations, including law firms. The pattern: in through edge devices, pivot to vCenter, then make the hypervisor quiet.</p>

<p>For responders, the lesson is simple. If your hypervisor investigation depends on logs that live on the hypervisor, you're counting on the attacker to preserve your evidence. Beyond this CVE, we see this all the time where the threat actor has SSH'd to your ESXi hosts and encrypted your VMDK's locally in their data store and also cleaned up after themselves by encrypting all the log files.</p>

<p>VMWare shops should be forwarding ESXi and vCenter logs off-host to a central store that virtualization admins can't modify, kept at least 90 days, and alert when forwarding stops. Pair it with lockdown mode and a management network that isn't reachable from user subnets. Be able to answer "who touched this host, and when" from data the attacker never had access to.</p>

<p>And patch the "lows" that land in KEV on the same timeline as the criticals.</p>

<p>#DFIR #VMware #ThreatHunting</p>

<p>https://www.securityweek.com/chinese-hackers-quietly-used-low-severity-esxi-logging-flaw-as-zero-day/</p>]]></content><author><name>Richard Brackett</name></author><category term="tactical" /><category term="vmware" /><category term="esxi" /><category term="logging" /><summary type="html"><![CDATA[A "low-severity" CVE just reminded me that severity scores are not threat models.]]></summary></entry><entry><title type="html">VPN Placement for SMBs</title><link href="https://securityvolition.com/blog/2026/11/07/vpn-placement-for-smbs/" rel="alternate" type="text/html" title="VPN Placement for SMBs" /><published>2026-11-07T09:15:00-05:00</published><updated>2026-11-07T09:15:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/07/vpn-placement-for-smbs</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/07/vpn-placement-for-smbs/"><![CDATA[<p>A SonicWall VPN bug this week is a good reminder: plan for the day your VPN gateway is the problem. Especially with SMB's, a TA connection to the VPN is an all-experience ticket to the network.</p>

<p>CVE-2026-19868 lets someone holding valid credentials for an alternate login method skip SAML on SonicOS SSL-VPN, potentially including the MFA you enforced there. SonicWall says it was exploited, and CISA added it to KEV on Nov 6. Fix: Gen7 7.3.4-7013 or Gen8 8.2.3-8015. No workaround so far. Patch first, then review VPN logins that never touched your identity provider.</p>

<p>Patching fixes the flaw. Architecture decides how much an attacker gets when the next one lands.</p>

<p>Cisco's 2016 SAFE remote access VPN design terminates the VPN at the edge, then puts a separate next-gen firewall and IPS in line "for inspection of the remote users sessions after tunnel termination." Translation: decrypted VPN traffic is still untrusted and you get full security boundary logging for what's hitting your LAN via the VPN.</p>

<p>I would adapt that for a small shop like this (my take, not Cisco's):</p>
<ul>
  <li>Put the VPN pool in its own zone. Treat it like a DMZ, not an extension of the LAN.</li>
  <li>Enforce policy at a boundary the VPN box doesn't control. VLANs on the same appliance aren't independent: if the gateway is owned, so are its rules. An existing internal firewall works if it's truly separate. If not, a small second firewall is often within reach financially. It just adds a little complexity.</li>
  <li>Allowlist only required destinations and services. No broad LAN access; permit AD-dependent traffic only where the application requires it, and keep management interfaces out of reach.</li>
  <li>Admin access goes through one hardened jump host with its own MFA, never straight to switches, hypervisors, or the firewall UI.</li>
  <li>Keep MFA, just don't make it the only control.</li>
  <li>Send VPN, IdP, and firewall logs somewhere the appliance can't erase.</li>
</ul>

<p>If most users only need two or three apps, consider app-scoped access instead of broad network access.</p>

<p>None of this stops an auth bypass. It shrinks what the bypass buys and if you are monitoring correct, you can alert on unusual traffic that your day to day users don't generate.</p>

<p>https://www.cisa.gov/news-events/alerts/2026/11/06/cisa-adds-one-known-exploited-vulnerability-catalog<br />
https://www.bleepingcomputer.com/news/security/sonicwall-warns-of-new-ssl-vpn-mfa-bypass-flaw-exploited-in-attacks/<br />
https://www.cisco.com/c/dam/en/us/solutions/collateral/enterprise/design-zone-security/safe-design-guide-edge-remote-access-vpn-ddos.pdf</p>

<p>#DFIR #IncidentResponse #NetworkSecurity</p>]]></content><author><name>Richard Brackett</name></author><category term="edge-devices" /><category term="vpn" /><category term="network-architecture" /><category term="sonicwall" /><summary type="html"><![CDATA[A SonicWall VPN bug this week is a good reminder: plan for the day your VPN gateway is the problem. Especially with SMB's, a TA connection to the VPN is an all-experience ticket to the network.]]></summary></entry><entry><title type="html">LLMjacking Bills Are Incident Signals</title><link href="https://securityvolition.com/blog/2026/11/06/llmjacking-bills-are-incident-signals/" rel="alternate" type="text/html" title="LLMjacking Bills Are Incident Signals" /><published>2026-11-06T10:08:00-05:00</published><updated>2026-11-06T10:08:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/06/llmjacking-bills-are-incident-signals</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/06/llmjacking-bills-are-incident-signals/"><![CDATA[<p>A stolen cloud key used to mean a slow cryptomining bill. Now it can mean a near seven-figure AI bill before lunch.</p>

<p>Google Threat Intelligence Group and Mandiant describe 2026 LLMjacking incidents where attackers used stolen access to run AI models at the victim's expense. In one case, more than 500,000 model calls in under eight hours left the victim with nearly a million dollars in token costs. The way in usually wasn't exotic: a misconfigured service account key, a GitHub token, or an active session token. In several cases the AI services were switched off until the attacker turned them on. Would you see that as an alert?</p>

<p>For responders, that changes the reaction clock. Billing is now an incident signal, and containment has to happen in minutes, not at the next cost review.</p>

<p>Everyone should treat AI services as a guarded resource. Keep them disabled in accounts that don't need them, alert on any change that enables a model or raises a quota, and replace long-lived API keys with short-lived, scoped credentials. Good looks like on-call getting paged within minutes of a model being enabled or a spend spike, plus a rehearsed playbook to revoke keys and cap quotas. Wouldn't this make a great table top exercise?</p>

<p>Your cloud bill is now part of your detection stack. Make sure someone is watching it.</p>

<p>#IncidentResponse #CloudSecurity #AISecurity</p>

<p>https://www.helpnetsecurity.com/2026/11/03/llmjacking-stolen-ai-access-cloud-bills/</p>]]></content><author><name>Richard Brackett</name></author><category term="cloud-ir" /><category term="ai-security" /><category term="llmjacking" /><summary type="html"><![CDATA[A stolen cloud key used to mean a slow cryptomining bill. Now it can mean a near seven-figure AI bill before lunch.]]></summary></entry><entry><title type="html">Alpine GitLab IR Write-up</title><link href="https://securityvolition.com/blog/2026/11/04/alpine-gitlab-ir-write-up/" rel="alternate" type="text/html" title="Alpine GitLab IR Write-up" /><published>2026-11-04T12:15:00-05:00</published><updated>2026-11-04T12:15:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/04/alpine-gitlab-ir-write-up</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/04/alpine-gitlab-ir-write-up/"><![CDATA[<p>One of the best incident write-ups I've seen this month came from Alpine Linux, after an attempted exploit of their GitLab server (CVE-2026-97229).</p>

<p>What they got right, from someone who delivers IR findings to customers:</p>

<ol>
  <li>
    <p>They said the hard thing plainly: "We have found no evidence that it succeeded, but we cannot rule it out, so we treated the instance as compromised and rebuilt it from scratch."</p>
  </li>
  <li>
    <p>They separated what they know from what they can't see due to logging gaps. Packages and releases were checked against records the attacker couldn't reach. They also named the gaps: some logs from the attack day are gone, and outbound connections weren't logged.</p>
  </li>
  <li>
    <p>They explained why their own upgrade didn't close the hole. GitLab's fix ships as a patch to the Ruby bundled in GitLab's packages, and their image used standard Ruby.</p>
  </li>
</ol>

<p>That's the bar I hold for customer briefings. State the confidence level, show the evidence behind it, say what you can't see, and make the decision the uncertainty calls for. Customers can handle "we can't rule it out." Let privacy counsel advise on how to handle that uncertainty. What costs trust is certainty that proves to be false later.</p>

<p>Practical check: if you run GitLab on your own Ruby or container image, you also need Ruby 3.3.13, 3.4.12, 4.0.9 or later.</p>

<p>https://www.alpinelinux.org/posts/GitLab-security-incident-CVE-2026-97229.html</p>

<p>#IncidentResponse #DFIR</p>]]></content><author><name>Richard Brackett</name></author><category term="dfir" /><category term="gitlab" /><category term="leadership" /><summary type="html"><![CDATA[One of the best incident write-ups I've seen this month came from Alpine Linux, after an attempted exploit of their GitLab server (CVE-2026-97229).]]></summary></entry><entry><title type="html">Ivanti Patching Isn&apos;t Eviction</title><link href="https://securityvolition.com/blog/2026/11/02/ivanti-patching-isnt-eviction/" rel="alternate" type="text/html" title="Ivanti Patching Isn&apos;t Eviction" /><published>2026-11-02T07:40:00-05:00</published><updated>2026-11-02T07:40:00-05:00</updated><id>https://securityvolition.com/blog/2026/11/02/ivanti-patching-isnt-eviction</id><content type="html" xml:base="https://securityvolition.com/blog/2026/11/02/ivanti-patching-isnt-eviction/"><![CDATA[<p>Patching an edge appliance is not the same as evicting an attacker.</p>

<p>The Ivanti Connect Secure chain that Volexity tracks as UTA0576 illustrates just why that is. Ivanti's Oct 13 fix closed the authentication bypass. The privilege escalation and file-upload flaws in the same chain were fixed later, and CISA added both to KEV on Oct 30.</p>

<p>What stands out from the IR perspective is that the actor patched code in memory, restored the configuration, deleted entries from the debug and audit logs, and left no web shell. Volexity says Ivanti's Integrity Checker doesn't detect that, so a clean ICT result is meaningless.</p>

<p>They also went after LAPS passwords and BitLocker recovery keys through the appliance's AD access. The incident doesn't stay on the VPN.</p>

<p>If you ran an affected version this month:</p>
<ul>
  <li>Upgrade to a fixed release (25.1.1.3 or 22.8R2.6)</li>
  <li>Scope with network and memory forensics, not file integrity alone (you are capturing netflows, right?)</li>
  <li>If compromise is suspected, follow CISA's recovery steps: factory reset or rebuild, restore the config (you have a backup, right?), reset the domain, admin and local passwords, reissue new certificates and keys</li>
  <li>Review who can read your LAPS and BitLocker data (which you have documented, right?)</li>
</ul>

<p>Edge devices are often where evidence is thinnest and blast radius is largest. Check out Cisco's reference architecture for placing VPN devices behind control points to enable network logging of VPN traffic: https://www.cisco.com/c/en/us/solutions/collateral/enterprise/design-zone-security/safe-secure-edge-architecture-guide.html</p>

<p>CISA's updated guidance: https://www.cisa.gov/news-events/alerts/2026/10/13/cisa-urges-patching-newly-exploited-vulnerabilities-ivanti-connect-secure</p>

<p>#DFIR #IncidentResponse #ThreatIntel</p>]]></content><author><name>Richard Brackett</name></author><category term="dfir" /><category term="ivanti" /><summary type="html"><![CDATA[Patching an edge appliance is not the same as evicting an attacker.]]></summary></entry></feed>