OpenAI Just Shipped a Model That Hacks for a Living
OpenAI released GPT-6 Cyber, a specialized model rated 'Critical' for cybersecurity. Its entire job is to find software vulnerabilities, write working exploits, and verify the fixes - and access is locked behind a vetted
OpenAI released GPT-6 Cyber, a specialized model rated 'Critical' for cybersecurity. Its entire job is to find software vulnerabilities, write working exploits, and verify the fixes - and access is locked behind a vetted red-team program.
Why it matters
- 948 working Linux kernel exploits found (588 previously unknown)
- 81 unknown high-severity Firefox bugs discovered; Mozilla is patching
- 304 critical findings disclosed to maintainers
- The whole run cost about 1.2M dollars in API credits
- It stayed in scope 99.93% of the time, up from 94.49% for the prior model
The exciting read: defenders get a tireless security teammate. The terrifying read: offensive capability is now a waitlisted API call, not a rare human skill.
Full story: https://www.threerouter.com/blog/332
Originally published by Dev.to AI. Aggregated on AIWithGhost for educational purposes — full credit and traffic to the original publisher.