OpenAI dangles $50K to break GPT-5.6's biosafety wall

OpenAI just doubled its bio bug bounty to $50,000 — up from $25K — and folded it into an ongoing private program targeting GPT-5.6. The ask is narrow but brutal: find one universal jailbreak prompt that answers all five of OpenAI's bio-safety challenge questions from a clean chat, no moderation trips.

This isn't an open free-for-all. OpenAI is inviting vetted red-teamers with AI security or biosecurity chops. Selected researchers sign an NDA and work through OpenAI's bounty platform — every prompt, completion, and finding stays under wraps.

The program keeps honoring the older GPT-5.5 scope until testing wraps July 27, 2026, after which only GPT-5.6 is in play. The bigger signal: as frontier models inch toward genuinely dangerous bio capabilities, OpenAI is betting that paying hackers to break safeguards first is cheaper than the alternative.

Why it matters: a $50K price tag on a single jailbreak is OpenAI admitting its biosafety guardrails are only as good as the people trying to break them.