OpenAI Doubles Bio Bounty to $50K, Makes Program Ongoing for GPT-5.6

OpenAI is upgrading its GPT-5.5 Bio Bug Bounty program into an ongoing private effort—the OpenAI Bio Bounty Program—with a renewed focus on finding universal jailbreaks that bypass biosafety guardrails. The core tension is clear: as frontier models become more capable in domains like biology, the risk of malicious use increases, but measuring and incentivizing real-world safety coverage is notoriously hard. The article surfaces this specific challenge by narrowing the bounty to defeating a predefined biosafety challenge, rather than vague or hypothetical harm scenarios.

The concrete changes are twofold. First, the program shifts from a fixed-term bug bounty to a rolling private program starting with GPT-5.6, making it a standing safety research effort rather than a one-time competition. Second, the top reward for a universal jailbreak is doubled from $25,000 to $50,000 for both GPT-5.5 and GPT-5.6, with smaller partial wins eligible at OpenAI‘s discretion. The scope for the GPT-5.5 bounty remains active until July 27, 2026; after that, only GPT-5.6 is in scope. Applicants must have an existing ChatGPT account and sign an NDA, but past applicants don’t need to reapply.

For builders and safety researchers, the takeaway is that OpenAI is treating biosafety jailbreaks as a persistent engineering problem, not a one-off audit. The move to a rolling private program suggests the evaluation methodology (the “predefined biosafety challenge”) is increasingly stable, and the doubling of the reward signals that the bar for a true universal jailbreak is high and worth serious effort. Anyone working on red-teaming or model safety should note the closing of the GPT-5.5 window and the explicit focus on universal (not narrow) jailbreaks as the primary reward target.