OpenAI: first Bio Bug Bounty — researchers invited to find biosecurity vulnerabilities in the GPT-5.5 model

On July 9, 2026, OpenAI launched a Bio Bug Bounty for GPT-5.5 — a program in which external researchers attempt to find biosecurity vulnerabilities in the model. A bug bounty is a practice borrowed from cybersecurity: a company pays independent researchers for responsibly disclosed vulnerabilities, rather than having them discovered by malicious actors.

Why does biosecurity get its own bounty?

Biological risks are considered the most sensitive category of frontier model misuse: a system with doctoral-level biology knowledge could lower the knowledge barrier for developing biological threats. Unlike ordinary jailbreaks, where the harm is mostly reputational, a biosecurity vulnerability has potentially catastrophic consequences — which is why labs treat this domain under separate, stricter standards. Anthropic activated ASL-3 protections for the same risk category last year alongside its own bounty for universal jailbreaks; OpenAI now formalizes external testing through the first bounty dedicated exclusively to biology.

What we know, and what we don’t

The full article returns a 403 for automated access, so reward amounts, eligibility criteria, and program duration remain unknown. What is clear is that the target is GPT-5.5 — the model that remains in widespread use even after yesterday’s launch of the GPT-5.6 family.

A signal to the industry

External crowdsourced red-teaming for the most dangerous risk category is becoming the standard: internal teams cannot cover all attack vectors, and a formal bounty gives ethical researchers a legal channel and motivation. After bug bounties became the norm in software security, the AI industry is now applying the same logic to the security of models themselves.