OpenAI's GPT‑6 Finds 81 Firefox Zero‑Days

OpenAI announced GPT‑6 Cyber, a restricted security model that markedly outperformed its predecessor on an exploit benchmark (48.5% vs 32.1%) and that OpenAI says helped identify 81 previously unknown high‑severity Firefox vulnerabilities subsequently noted in a Mozilla advisory. The claim is paired with controlled tests of scope compliance but does not prove safe autonomous deployment in the wild. — If accurate, AI‑scale vulnerability discovery shifts the balance between offensive capability and defensive patching, forcing new norms, regulation, and institutional responsibility for disclosure and use of automated exploit tools.