Krux
April 16, 2026
Anthropic's Secret Model Just Found a 27-Year-Old Bug
Published: April 16, 2026 at 12:30 AM
Updated: April 16, 2026 at 12:30 AM
100-word summary
Anthropic is giving Apple, Microsoft, Google, and 40+ partners early access to Claude Mythos Preview, a frontier model you can't use, to hunt zero-day vulnerabilities before attackers do. The model already uncovered a 27-year-old OpenBSD flaw and a 16-year FFmpeg bug. Partners get up to $100M in credits to scan their own code and open-source projects, with all findings public within 90 days. The catch? Mythos is too dangerous for general release. Anthropic is betting that giving the good guys advanced AI first, under tight controls, beats waiting for hackers to build their own version.
What happened
Anthropic is giving Apple, Microsoft, Google, and 40+ partners early access to Claude Mythos Preview, a frontier model you can't use, to hunt zero-day vulnerabilities before attackers do. The model already uncovered a 27-year-old OpenBSD flaw and a 16-year FFmpeg bug. Partners get up to $100M in credits to scan their own code and open-source projects, with all findings public within 90 days. The catch? Mythos is too dangerous for general release.
Why it matters
Anthropic is betting that giving the good guys advanced AI first, under tight controls, beats waiting for hackers to build their own version.