GPT-5.5 Bio Bug Bounty

Source: OpenAI Date: 2026-04-23 URL: https://openai.com/index/gpt-5-5-bio-bug-bounty

Summary

Companion launch to GPT-5.5 (April 23, 2026), OpenAI’s bio-specific bug bounty program inviting biosecurity researchers to probe GPT-5.5 for unsafe biological uplift — information the model might provide that could enable harm in a wet-lab setting. The program follows the pattern of the earlier safety bug bounty (March 2026) but narrows scope specifically to bio risks, which were flagged in the GPT-5.5 system card as a primary concern given the model’s reasoning capability improvements.

Implications

Bio risk as the frontier safety concern. The fact that OpenAI launched a bio-specific bounty alongside GPT-5.5 — not a generic safety bounty, not a general red-team program — signals that internal evals flagged biosecurity uplift as the highest-concern risk category for this model tier. That’s a meaningful public acknowledgment.

Responsible disclosure playbook. The bio bug bounty follows the same structure as the broader safety bug bounty: external researchers, defined scope, payout structure. It’s OpenAI operationalizing a safety-community relationship at the frontier. Whether the payout scale is sufficient to attract serious biosecurity expertise is a separate question.

Thread: GPT-5.5 launch cluster. This signal is one of four simultaneous April 23 launches (GPT-5.5 intro, top 10 Codex uses, Codex settings, bio bounty) — OpenAI releasing safety documentation alongside product launches rather than after criticism, which is the model spec era playbook.

Watch: Whether independent biosecurity orgs (NTI, Johns Hopkins Center for Health Security) engage with the program or view it as insufficient.