TLDR: OpenAI has launched a specialized bio bug bounty program for its new ChatGPT Agent, inviting expert researchers to identify and report vulnerabilities related to biological and chemical risks. The initiative aims to enhance the safety protocols of the advanced AI model by challenging participants to find ‘universal jailbreaks’ that can bypass its ten-level bio/chem safety system, with significant monetary rewards for successful exploits.
OpenAI has announced the launch of a targeted bio bug bounty program, specifically designed to bolster the safety and security of its recently deployed ChatGPT Agent. This proactive measure underscores the company’s commitment to mitigating potential biorisks associated with advanced artificial intelligence capabilities. The program, which opened for applications on July 17, 2025, with testing slated to commence on July 29, 2025, seeks to engage the global research community in identifying critical vulnerabilities.
The core challenge of the bounty program is for researchers to discover a ‘universal jailbreaking prompt’ capable of successfully answering all ten bio/chem safety questions from a clean chat session with the ChatGPT Agent. This rigorous test is designed to evaluate the robustness of the model’s current safety mechanisms against sophisticated attempts to bypass its safeguards. OpenAI has classified the ChatGPT Agent as having ‘high bio/chemical risk,’ necessitating such a focused security initiative.
Substantial rewards are being offered for successful findings. The first researcher or team to achieve a true universal jailbreak that unequivocally clears all ten challenge questions will receive $25,000. Additionally, a reward of $10,000 is designated for the first team that manages to answer all ten questions using multiple distinct jailbreak prompts. OpenAI also reserves the right to grant smaller awards for partial successes, acknowledging valuable contributions that may not fully meet the top-tier criteria.
Participation in the program is strictly by application and invitation only. OpenAI is actively vetting trusted bio red-teamers and reviewing new applications. Prospective participants are required to have an existing ChatGPT account and must submit a concise application by July 29, 2025, including a brief track record and a 150-word plan outlining their approach. All prompts, completions, findings, and communications within the program will be covered by a Non-Disclosure Agreement (NDA), ensuring confidentiality.
Also Read:
- OpenAI’s New ChatGPT Agent Poses Significant Bioweapon and Financial Misuse Risks
- OpenAI’s New ChatGPT Agent Demonstrates Autonomous Capabilities, Navigating Real-World Tasks with Noted Speed Limitations
The ChatGPT Agent itself represents a significant evolution, transforming the popular chatbot into a personal AI assistant capable of handling multi-step tasks, such as web browsing, data analysis, and presentation creation, by operating its own virtual computer. OpenAI states that this model incorporates its ‘most comprehensive safety stack to date,’ involving extensive threat modeling, monitoring, and collaboration with external biosecurity experts for red-teaming prior to its launch. Despite these precautions, OpenAI acknowledges that ‘with this model there are more risks than with previous models,’ emphasizing the need for continuous refinement and adaptive safeguards as the AI operates in real-world scenarios. This bug bounty is a critical step in their ongoing efforts to ensure frontier AI systems are developed and deployed safely and securely.
