OpenAI doubles its bounty for biological jailbreaks to $50,000

On July 9, 2026, OpenAI moved the GPT-5.5 Bio Bug Bounty into a permanent, closed format and doubled the maximum reward for a universal bypass of the models' biological restrictions, from $25,000 to $50,000.

How the new program works

OpenAI renamed the initiative the OpenAI Bio Bounty Program and made it permanent. Researchers try to find universal jailbreaks: ways to get around a predefined set of biosafety restrictions that work not in a single one-off prompt but systematically against the company's models.

OpenAI's frontier models will be tested, starting with GPT-5.6. The goal is to find weaknesses in safety measures before attackers can exploit them to obtain dangerous biological information or instructions.

• On July 9, 2026, OpenAI announced the launch of the permanent, closed OpenAI Bio Bounty Program.
• The reward for a universal safeguard bypass for GPT-5.5 and GPT-5.6 rose from $25,000 to $50,000.
• Testing of GPT-5.5 under the previous program ends on July 27, 2026.
• After July 27, 2026, GPT-5.6 remains in scope.
• For partial results, OpenAI may award smaller payouts at its discretion.

The program is not a public contest where you can immediately submit a bypass you have found. Applicants must submit a short application with their name, affiliation and a description of their experience. Selected participants will be given access to a dedicated testing platform. Researchers who already applied to the GPT-5.5 Bio Bounty do not need to reapply.

What will researchers look for?

Researchers will test whether a single reproducible method can remove the set restrictions tied to biological risks. OpenAI calls such findings universal jailbreaks: not a single poorly worded prompt, but a vulnerability that can consistently bypass the model's safeguards.

The company has not disclosed the technical conditions of its set of biosafety tasks. That is expected: publishing exact criteria and examples could make it easier to develop bypasses outside the controlled program. Participants will work in a closed environment, and they need an active ChatGPT account and must sign a non-disclosure agreement.

The higher payout gives model safety evaluators a markedly stronger incentive to hunt for such problems. At the same time, the closed format lets OpenAI vet participants' qualifications, control the testing environment and process results before details of a potential vulnerability become public.

Why the focus on GPT-5.6

After July 27, 2026, the program will stop accepting results for GPT-5.5 and switch to GPT-5.6. OpenAI stated explicitly that this model will be the starting point for ongoing testing, and that it will notify researchers separately about future changes to the scope.

The shift shows that biosafety testing will accompany the release of new frontier models rather than remain a one-off campaign around GPT-5.5. For OpenAI, it is a way to get outside scrutiny of its safeguards: the internal team defines the protective perimeter, and independent specialists look for repeatable ways around it.

The Bio Bounty program exists separately from OpenAI's other bug-finding initiatives. The company also directs researchers to its general security vulnerability program and to the Security Bug Bounty program. The new format focuses specifically on risks that arise when advanced AI models are applied in biology.

What this means

OpenAI is raising the price for finding the most dangerous bypasses and turning biological testing into a continuous process. For the industry, it is a signal: risk assessment of frontier models is increasingly an ongoing external review, not a phase that ends with the release.