A 17-year-old FreeBSD vulnerability. Unauthenticated root from the open internet. Discovered and fully exploited — without a human in the loop.
This is not a red team exercise. This is what Anthropic's Claude Mythos Preview did in production evaluation.
Mythos fully autonomously identified the flaw, developed a working exploit, and demonstrated that any unauthenticated attacker anywhere on the internet could gain root access to affected servers. The vulnerability had survived 17 years of code review, fuzzing campaigns, and manual security audits.
FreeBSD NFS remote code execution, CVE-2026-4747, was 17 years old. Mythos built a 20-gadget ROP chain split across multiple packets. Fully autonomous. The Linux kernel local privilege escalation chained two to four low-severity vulnerabilities into full local privilege escalation via race conditions and KASLR bypasses.
The governance gap: No existing AI policy framework — not NIST AI RMF, not the EU AI Act, not CISA guidance — addresses what happens when a frontier model discovers thousands of weaponizable zero-days before patches exist. The disclosure window is no longer measured in days. It's measured in model inference time.
What defenders must do now:
1. Patch FreeBSD NFS exposure immediately — CVE-2026-4747 is public.
2. Assume your threat model is now priced at AI inference costs, not human researcher time.
3. Pressure your AI governance teams to define policy on AI-discovered zero-day obligations — before a breach forces the question.
♾ The AI Threat Brief | AI Security Intelligence for Leaders
