OpenAI is enlisting external researchers to probe for biological risks within its forthcoming GPT-5.5 model. The company announced a "Bio Bug Bounty" program, inviting experts in AI red teaming, security, and biosecurity to identify vulnerabilities.
The core challenge is to discover a single, universal jailbreaking prompt that can bypass OpenAI's five-question bio-safety challenge across the GPT-5.5 model, specifically within the Codex Desktop environment. This initiative reflects a growing effort to bolster safeguards for advanced AI capabilities, particularly those with potential biological implications, as detailed in OpenAI's research announcement.
The Bio Bug Bounty Program
The program, which opens for applications on April 23, 2026, offers a $25,000 reward for the first researcher to successfully devise a universal jailbreak that circumvents all five bio-safety questions without triggering moderation. Partial wins may also be awarded at OpenAI's discretion.
Applications are open until June 22, 2026, with testing scheduled from April 28 to July 27, 2026. Selected participants will gain access to the bio bug bounty platform, operating under a strict non-disclosure agreement covering all prompts, findings, and communications. This program is a critical step in understanding and mitigating potential dual-use risks inherent in powerful AI, a concern amplified by advancements like those seen in OpenAI's GPT-5.5.
The search for such vulnerabilities highlights the ongoing need for rigorous testing, including exploring the limits of models through techniques like finding a universal jailbreak AI.