OpenAI releases GPT-6 Cyber, rated Critical for cyber capability
Original title: OpenAI releases GPT-6 Cyber, and promises this one stays within 'rules of engagement'
Access starts with approved Daybreak Red customers after OpenAI said the model stayed in scope over 99 percent of the time.
OpenAI released GPT-6 Cyber, a cybersecurity model it says finds and exploits vulnerabilities better than GPT-5.6 Cyber and GPT-6 Astra while staying inside authorized scope. On ExploitGym it scored 48.5 percent versus 32.1 percent for GPT-5.6 Cyber. Researchers used it to report 81 new high-severity Firefox bugs and working exploits for 948 Linux kernel vulnerabilities, including 588 previously unknown. It is initially limited to approved Daybreak Red API and Codex customers.
Unconfirmed: OpenAI did not detail the 0.07 percent of cases that exceeded scope or confirm any real incidents.
- Only 0.07 percent of boundary tests went beyond the assigned task scope.
- Mozilla has started fixing 81 previously unknown high-severity Firefox flaws.
- Linux findings, with potential patches, were sent to maintainers for review.
- OpenAI will later widen access and also offer GPT-5.6 Cyber to Daybreak Red customers.
- Critical bug-bounty payouts on core products now reach up to $30,000 before bonuses.
From the original article
2 min read at theregister.comThe latest bug hunting model 'avoided exploiting out-of-scope services over 99 percent of the time', claims House of Altman
By Jessica Lyons · theregister.com
A short excerpt; the article and its images belong to the publisher. Summary, key facts and assessment above are AI-written from the article and can be wrong.