OpenAI Bio Bounty Goes Ongoing: $50K Jailbreaks as Launch-Day Infrastructure
Same day GPT-5.6 went public, OpenAI turned its GPT-5.5 bio bounty into an ongoing private Bio Bounty Program—universal jailbreaks vs a fixed biosafety challenge, top award $25K→$50K, GPT-5.6 in scope. External adversarial testing as continuous ops, not a side quest.
Model launch days usually bury safety ops. Pairing a public GPT‑5.6 unlock with a doubled, permanent bio jailbreak bounty is OpenAI advertising external adversarial testing as continuous infrastructure—especially for a risk domain (biorisk) that governments and enterprises treat as non-negotiable.
OpenAI on July 9, 2026 announced that its GPT‑5.5 Bio Bug Bounty is becoming an ongoing private program—the OpenAI Bio Bounty Program—aimed at universal jailbreaks that defeat a fixed biosafety challenge on frontier models, starting with GPT‑5.6 and continuing forward. The top reward rises from $25,000 to $50,000 for a qualifying universal jailbreak on GPT‑5.6 and GPT‑5.5; partial awards remain discretionary. GPT‑5.5 stays in scope through the prior testing window (through July 27, 2026 per contemporaneous program notes).
What changed#
From OpenAI’s primary “OpenAI Bio Bug Bounty” safety post (July 9) and TechRepublic / program page context:
| Item | Detail |
|---|---|
| From | Time-boxed GPT‑5.5 bio red-team bounty (applications earlier in 2026) |
| To | Ongoing private Bio Bounty Program |
| Target | Universal jailbreaks vs predefined biosafety challenge |
| Models | GPT‑5.6 in scope from launch; GPT‑5.5 retained |
| Top award | $50,000 (was $25,000) |
| Access | Invite / application-based private program for vetted red-teamers |
OpenAI frames the program as part of strengthening safeguards for advanced biology capabilities as frontier models ship—explicitly tying the escalation to the GPT‑5.6 public era. It also sets a competitive benchmark: other labs either match external bio red-team incentives or explain why not.
Limits#
- Private/invite program—results and award rates are not public dashboards.
- “Universal jailbreak” criteria are OpenAI-defined against a fixed challenge; not a general bio-capability eval suite.
- Partial awards remain discretionary.
Sources#
- OpenAI: “OpenAI Bio Bug Bounty” (July 9, 2026)
- OpenAI prior GPT‑5.5 Bio Bug Bounty materials (April 2026) for baseline $25K structure.
- TechRepublic and program application page notes on $50K raise and GPT‑5.6 scope (July 9–10, 2026).