OpenAI is running a paid red-teaming challenge on GPT-5.5 targeting biosafety jailbreaks, with up to $25,000 for a universal exploit.
Key Takeaways
The top prize ($25,000) requires finding a single universal jailbreak that clears all five biosafety test questions.
Participants must have a ChatGPT account and sign an NDA before accessing program details.
Applications ask for a proposed jailbreak approach before revealing what the actual test questions are.
Access is invite-only: OpenAI will extend invitations to a vetted list of trusted bio red-teamers.
Hacker News Comment Review
Commenters flagged a structural problem: applicants must describe their jailbreak approach before seeing the questions, making a coherent proposal nearly impossible.
The NDA requirement drew skepticism as a silencing mechanism that prevents public disclosure of findings, undermining the credibility of the program as genuine security research.
The winner-takes-all payout structure means only one researcher collects the $25,000 regardless of how many valid exploits are independently discovered, which commenters called a scam framing.
Notable Comments
@applfanboysbgon: “Even if 100 people find ‘bugs’, they will only pay out to one person” – winner-takes-all framing undercuts the bounty premise entirely.