OpenAI Offers Up to $25,000 for Jailbreaks of GPT-5.5 Biosafety Guardrails

OpenAI has launched a bug bounty program offering rewards of up to $25,000 to security researchers who can find universal jailbreaks capable of bypassing biosafety guardrails in its GPT-5.5 model, the company announced on its blog.

The GPT-5.5 Bio Bug Bounty targets vulnerabilities specifically in the biological risk domain, according to OpenAI. The program seeks red-teamers who can demonstrate repeatable methods for circumventing safety protections designed to prevent the model from providing dangerous biological information.

The initiative focuses on “universal jailbreaks” — techniques that reliably bypass safety filters rather than one-off prompt manipulations, OpenAI said. The distinction signals the company’s interest in identifying systemic weaknesses in its safety architecture rather than isolated edge cases.

Bug bounty programs have become standard practice in the software industry, but their application to AI biosafety represents a relatively new approach to model security. Several AI providers have adopted red-teaming programs in recent years as frontier models grow more capable and regulators increase scrutiny of potential misuse risks.

The bounty arrives as policymakers in the United States and European Union continue to develop frameworks for AI safety oversight. The Biden administration’s 2023 executive order on AI and the EU AI Act both emphasized the importance of red-teaming and adversarial testing for high-capability AI systems.

OpenAI has previously operated bug bounty programs through third-party platforms for its products, but the GPT-5.5 Bio Bug Bounty narrows the scope to biological risk — an area that has drawn particular attention from national security officials and biosafety researchers concerned about AI-enabled threats.

The $25,000 maximum payout places the program at the higher end of AI-specific bounty rewards, reflecting the severity OpenAI assigns to potential biosafety failures in its latest model.

OpenAI, based in San Francisco, develops the GPT series of large language models and the ChatGPT consumer product. The company has faced ongoing debate over how to balance the rapid deployment of increasingly powerful AI systems with adequate safety testing.

Source: OpenAI Blog