Posted in Application Security

An AI pipeline just reproduced 1,724 Linux kernel vulnerabilities for about $1.2M in compute. Your patch…

Craig M. — Chief Product Officer @ Cyberr®

An AI pipeline just reproduced 1,724 Linux kernel vulnerabilities for about $1.2M in compute. Your patch cycle was not designed for that.

This week OpenAI described a run where an unreleased cyber model turned candidate kernel findings into reproduced bugs and working privilege escalation exploits, at industrial scale. The same week, OpenSSH shipped an emergency fix (10.6p2) for a password bypass on musl-based systems with PAM disabled, which covers Alpine's default sshd setup. Previdian reported a first exploitation attempt against its honeypot less than nine hours after the CVE went public.

Put those together and the lesson is uncomfortable: discovery and weaponisation are getting cheap and fast. Remediation is not. Most patch programmes still run on a monthly cadence and a CVSS cut-off, which quietly assumes attackers need weeks.

What I'd check this week:

→ Inventory Alpine/musl images that expose sshd. Most containers never needed it. Remove it rather than patch it.

→ Stop deprioritising local privilege escalation. If one web app foothold becomes root in hours, kernel patch SLAs for internet-facing hosts should look like your critical RCE SLAs.

→ Measure time-to-patch for exposed assets, not the average. The average hides exactly the hosts that matter.

→ Shrink what is reachable. Every service you switch off is one less race you have to win.

The hopeful part: defenders get these tools too. But only teams that have automated the boring middle (asset inventory, image rebuilds, reboot windows) will be able to absorb findings arriving at AI speed.

Is your patch process built for 30 days, or for 9 hours?

#VulnerabilityManagement #AISecurity #LinuxSecurity

3 reactions · 1 comments

Topics: #vulnerabilitymanagement, #aisecurity, #linuxsecurity

Comments

Victor Font: Craig, this exposes a fundamental mismatch between the speed of AI-assisted vulnerability discovery and the operational assumptions underlying traditional remediation programs. A 30-day patch cycle is not inherently a governance failure, but applying it indiscriminately to vulnerabilities that can be weaponized within hours certainly can be. The more important question is whether organizations can

Read and join the discussion on Cyberr