Krux
April 13, 2026
Anthropic's AI Found a 27-Year-Old OpenBSD Flaw
Published: April 13, 2026 at 12:31 AM
Updated: April 13, 2026 at 12:31 AM
100-word summary
Anthropic's Claude Mythos Preview has uncovered thousands of zero-day vulnerabilities, including bugs that survived 27 years in OpenBSD and 16 years in FFmpeg. The AI catches 83% of vulnerabilities in security tests, beating the previous Claude model's 67%. Here's the catch: you probably can't use it. Anthropic is limiting access to 12 launch partners plus about 40 critical-infrastructure groups, with no plans for public release. They're offering $100M in usage credits and charging partners $25 per million input tokens. The strategy seems designed to quietly stress-test whether AI can replace traditional security audits before the model finds its way into less careful hands.
What happened
Anthropic's Claude Mythos Preview has uncovered thousands of zero-day vulnerabilities, including bugs that survived 27 years in OpenBSD and 16 years in FFmpeg. The AI catches 83% of vulnerabilities in security tests, beating the previous Claude model's 67%. Here's the catch: you probably can't use it. Anthropic is limiting access to 12 launch partners plus about 40 critical-infrastructure groups, with no plans for public release. They're offering $100M in usage credits and charging partners $25 per million input tokens.
Why it matters
The strategy seems designed to quietly stress-test whether AI can replace traditional security audits before the model finds its way into less careful hands.