Recruiting for Agent Bio Bug Bounty

OpenAI is launching a bio bug bounty program to strengthen safeguards against biological and chemical risks in the ChatGPT agent. The program aims to find 'universal jailbreaks' that bypass security on questions where AI agent models could pose biological risks.

Participants will be tasked with passing all 10 stages of bio/chem safety questions using a single jailbreak prompt. The reward structure is as follows:

• $25,000: for the first successful universal jailbreak that passes all 10 questions
• $10,000: for the first team to solve all 10 questions using multiple jailbreak prompts
• Smaller rewards may be granted at OpenAI's discretion for other partial achievements.

Applications open on July 17, 2025, and the program operates on an invitation-only basis targeting verified red-team experts. Selected participants must sign an NDA (non-disclosure agreement), and all prompts and outputs will be treated as confidential.