Overview
- OpenAI announced Thursday that it rebranded its one‑off GPT‑5.5 bug bounty into an ongoing OpenAI Bio Bounty Program and raised the top reward to $50,000 to attract outside researchers.
- The program initially covers GPT‑5.6 and keeps GPT‑5.5 eligible for submissions through July 27, 2026, with smaller discretionary payments available for partial findings.
- OpenAI is asking vetted participants to produce a single reproducible 'universal jailbreak' prompt that reliably defeats the company’s predefined bio and chemistry safety questions without triggering moderation.
- Participation is by invitation or application, requires a ChatGPT account, a signed nondisclosure agreement, and use of OpenAI’s private testing platform, which limits what researchers may publicly disclose.
- OpenAI’s move follows its GPT‑5.5 system card that recorded sustained expert jailbreaking causing model‑level biosafety failures that were later blocked by the final safeguard stack, and it signals increased reliance on external, private red‑teaming to find repeatable risks before wider deployment.