OpenAI and Anthropic are in negotiations for a deal that would allow each company to stress-test the other’s AI systems to help identify safety risks, according to reports. The discussions center on testing methods designed to surface potential vulnerabilities before they can be exploited.
The New York Post, citing The Information and a person familiar with the talks, says the negotiations began earlier this year and involve a legally binding arrangement. It also links the timing of the talks to broader developments in the AI industry, including high-profile incidents mentioned in the reporting. NDTV focuses on the purpose of the negotiations—mutual stress-testing for safety—without detailing the legal status or specific events.
Taken together, the coverage indicates both companies are exploring a structured, reciprocal approach to evaluating AI behavior under challenging conditions. The extent of the legal framework and the specific incidents referenced are not fully detailed across the outlets, but both accounts describe ongoing talks.
