The same hacker AI found 81 Firefox flaws nobody knew about

The same model behind the Linux haul found 81 previously unknown Firefox vulnerabilities that Mozilla classified as high severity, sending reproducible evidence and proposed patches for each; Mozilla began shipping fixes shortly after. It also showed 30 known Firefox issues were worse than first reported.

The Firefox findings sit alongside 948 working Linux kernel privilege-escalation exploits built from 1,724 reproduced flaws, plus a critical Brotli bug rated 9.2 out of 10. OpenAI now rates GPT-6 Cyber Critical for cybersecurity — its highest preparedness tier — and it scored 48.5% on the ExploitGym benchmark, up from 32.1% for GPT-5.6 Cyber.

Access stays gated: only selected approved Daybreak Red customers get the model through the API and Codex, with extra safeguards for industrial controls, election systems and other high-risk targets. OpenAI says the model stayed within researcher-set boundaries in 99.93% of tests without safety prompts, and 100% with them.