OpenAI has launched a major competition for security researchers and developers to find previously unknown vulnerabilities in a new language model. The total prize pool is half a million dollars.

On August 5, 2025, OpenAI released new open-weight reasoning models called gpt-oss. Along with the release, the company announced a competition to find potential vulnerabilities and "malicious behavior" in one of the versions of the model, gpt-oss-20b.

Terms of Participation

Researchers and development teams have 21 days to submit their applications. Each team can consist of a maximum of five people and can submit only one application to participate in the competition.

Prize Pool and Awards

The 500 000 USD prize pool will be divided equally among the ten winning teams. Each finalist team will receive 50 000 USD. The winners will be determined by an expert committee consisting of OpenAI experts, as well as experts from other leading research labs.