OpenAI launches GPT-5 Bio Bug Bounty to test safety with universal jailbreak prompt

TL;DR

• OpenAI announces a Bio Bug Bounty inviting researchers to test GPT-5’s safety.
• Testing uses a universal jailbreak prompt as part of the evaluation.
• Prizes can reach up to $25,000 for qualifying findings.
• Details are available on the OpenAI page: https://openai.com/gpt-5-bio-bug-bounty

Context and background

OpenAI has published a call to researchers to participate in its Bio Bug Bounty program. The initiative centers on evaluating the safety of GPT-5 by using a universal jailbreak prompt as part of the assessment. The program offers monetary rewards for findings that meet the program’s criteria, with a maximum prize of $25,000. The announcement ties OpenAI’s ongoing safety and risk assessment efforts to a formal bug bounty framework, inviting external researchers to contribute to the evaluation process.

What’s new

This notification marks the launch of a dedicated Bio Bug Bounty focused on GPT-5. It formalizes an external testing channel for safety evaluation, specifying a method—the use of a universal jailbreak prompt—and a prize structure that includes awards up to $25,000 for verified submissions.

Why it matters (impact for developers/enterprises)

For developers and enterprises integrating AI products, this initiative underscores a public-facing commitment to safety testing and vulnerability disclosure. By inviting researchers to probe GPT-5’s safety with structured prompts, OpenAI aims to identify potential weaknesses and improve the model’s behavior in real-world usage. The program signals how large AI deployments may rely on external security testing as part of overall risk management.

Technical details or Implementation (as disclosed)

The core activity described is safety testing of GPT-5 using a universal jailbreak prompt. While the post does not disclose step-by-step testing procedures, it clearly positions the prompt as a tool for evaluating how GPT-5 handles challenging or misleading instructions. The exact submission guidelines, evaluation criteria, and workflow are presumably outlined on the official page linked below.

Key takeaways

• OpenAI has launched a Bio Bug Bounty for GPT-5 safety testing.
• The testing approach centers on a universal jailbreak prompt.
• Rewards for eligible findings can go up to $25,000.
• More details are available at the official OpenAI page.

References

• https://openai.com/gpt-5-bio-bug-bounty