OpenAI and Anthropic explore mutual model testing as security concerns rise

On September 21, 2026, The Information reported that OpenAI and Anthropic had discussed a legally binding agreement to conduct mutual stress tests on each other's AI models, ahead of a series of cybersecurity incidents involving OpenAI's technology. The proposed protocol aimed to identify vulnerabilities and risks through various tests. While it remains unclear if the agreement was finalized before recent security breaches, both companies had been in negotiations with their legal teams earlier this year. The idea of mutual testing mirrors Elon Musk's suggestion for AI labs to conduct security checks before commercial releases. OpenAI CEO Sam Altman supported a different approach, advocating for independent third-party assessments and industry-wide risk standards. A source close to OpenAI noted the company was exploring multiple security collaborations, including with government entities, and had expanded its focus to include pre-training and pre-release security measures. The proposed agreement would grant access to commercial AI models via API, without retaining any data from the other company.