OpenAI launches GPT-6 Cyber for in-scope vulnerability research and fixes
On November 5, OpenAI launched GPT-6 Cyber for finding, validating, and helping fix vulnerabilities inside the scope researchers set. It reports ExploitGym scores of 48.5% versus 32.1% for GPT-5.6 Cyber, and 100% in-scope behavior with its deployment prompt in their out-of-scope attack tests; access starts with approved Daybreak Red customers. My take: the stronger the digger, the more your first question is who may use it, how scope is locked, and who owns the fallout—not whether you can buy the tool. A fuzzy authorization letter is a red-team key left on the lobby desk.