Claude Wrote a FreeBSD Kernel Exploit in Four Hours — What AI-Powered Vulnerability Research Means for Security

Anthropic's Frontier Red Team had Claude Opus 4.6 autonomously develop two working remote root exploits for FreeBSD kernel flaw CVE-2026-4747 in roughly four hours. The same research pipeline has found 500+ high-severity vulnerabilities across major open-source projects. Here's what happened, how it works, and why it changes the security calculus for every team running production infrastructure.