OpenAI · 2026-04-23 · notable
OpenAI GPT-5.5 Bio Bug Bounty: $25K for a Universal Bio Safety Jailbreak
OpenAI is paying $25K for the first universal jailbreak of GPT-5.5's bio safety guardrails. Vetted researchers must find one prompt that bypasses all five biosafety questions without triggering moderation. Applications open April 23–June 22; testing April 28–July 27.
OpenAI pays $25K for the first prompt that bypasses GPT-5.5's bio safety questions — the first paid crowdsourced red-team for bio guardrails on a shipped frontier model.
What is it?
OpenAI launched a paid adversarial red-team challenge on April 23, 2026, targeting GPT-5.5's biological harm guardrails. Vetted participants — biosecurity researchers, AI red-teamers, and security professionals accepted through an application review — must find a single universal jailbreak prompt that makes GPT-5.5 answer all five biosafety challenge questions in a clean chat session without triggering moderation. Top prize: $25,000 for the first successful universal jailbreak. Applications close June 22, 2026; testing runs April 28–July 27. Participation requires NDA coverage of all findings.
How does it work?
The challenge is scoped to GPT-5.5 running in Codex Desktop only. A qualifying jailbreak must be universal — the same prompt must consistently bypass all five bio safety questions, not a session-specific or model-state-dependent exploit. OpenAI screens applications for relevant background (AI red-teaming, security, or biosecurity) and requires identity and organizational affiliation. Smaller discretionary awards may be given for partial successes judged at OpenAI's discretion.
Why does it matter?
Frontier models develop dangerous domain competence as they scale, and bio-harm guidance is among the highest-stakes misuse categories. OpenAI is specifically hardening the harder attack surface: a universal jailbreak that works across prompts and sessions, not just a one-off exploit. Paying external experts to find this before bad actors do is a meaningful shift in how labs approach safety evaluation on shipped models. This is the first formal, paid crowdsourced red-team targeting bio safety on a released GPT model.
Who is it for?
Biosecurity researchers, AI safety red-teamers, security professionals with relevant backgrounds
Try it
Apply at openai.smapply.org/prog/gpt-5-5-safety-bio-bounty-program/ (applications close June 22)