The AI giant is transitioning its biosecurity testing into a permanent, invitation-only program with beefed-up payouts for finding universal jailbreaks
OpenAI just made it significantly more lucrative to break its AI models, at least if you’re doing it for the right reasons. The company announced on July 9 that it’s overhauling its biosecurity bug bounty initiative, doubling the maximum reward for discovering universal jailbreaks from $25,000 to $50,000.
The revamped effort, now called the OpenAI Bio Bounty Program, replaces the earlier GPT-5.5 Bio Bug Bounty with something more permanent. Instead of one-off testing windows tied to specific model launches, OpenAI is building an ongoing infrastructure for catching the kinds of vulnerabilities that keep biosecurity experts up at night.
What’s actually changing #
The new program covers both GPT-5.6, OpenAI’s latest model, and ongoing evaluations of GPT-5.5 until that model’s original testing window closes on July 27, 2026. The $50,000 top-tier payout applies across both models, with OpenAI retaining discretion to issue partial awards for findings that don’t quite hit the universal jailbreak threshold but still surface meaningful risks.
Participants need to craft a single prompt that effectively answers all of OpenAI’s predefined bio and chem safety questions without triggering the model’s moderation systems.
The program operates strictly on an invitation or application basis. Interested researchers submit through a Google Form, and those accepted need a ChatGPT account, must sign a non-disclosure agreement, and get access to a private testing platform. OpenAI is specifically targeting experts in AI security and biosecurity.
The road from red teaming to permanent bounties #
OpenAI’s journey to this point has been incremental but deliberate. The company first dipped into targeted bio-risk red teaming back in July 2025, offering monetary incentives for uncovering system vulnerabilities. In 2025, OpenAI also ran a dedicated agent-focused bounty program. The GPT-5.5 Bio Bug Bounty launched on April 23, 2026, with applications closing on June 22.
What this means for the AI industry and investors #
The NDA requirement and private nature of the program reveal a tension at the heart of AI safety research. OpenAI wants external experts poking at its models, but it doesn’t want the specific vulnerabilities they find becoming public knowledge. That means the broader research community won’t have visibility into what’s being discovered or how quickly it’s being patched.
Disclosure: This article was edited by Editorial Team. For more information on how we create and review content, see our