GPT 5.5 biosafety bounty (openai.com)

A new initiative called the Bio Bug Bounty has been announced to enhance the safety measures surrounding advanced AI capabilities in biology, specifically targeting the GPT-5.5 model in Codex Desktop. The challenge invites researchers skilled in AI red teaming and biosecurity to uncover a universal jailbreak—a singular prompt capable of bypassing the model's safeguards while answering a series of five bio safety questions. The first successful contestant will earn a reward of $25,000, with smaller prizes available for partial successes. This announcement is significant for the AI/ML community as it underscores the growing concern for biosafety and the ethical implications of advanced AI capabilities. By actively soliciting the expertise of the research community, the initiative aims to identify potential vulnerabilities before they can be exploited. The challenge will run from April to July 2026, with a structured application process and strict nondisclosure agreements to maintain confidentiality. This effort not only reflects the industry's broader commitment to safety and accountability but also gives an opportunity for collaboration between AI developers and independent researchers to address potential risks associated with powerful AI technologies.