OpenAI and Anthropic reportedly discussed mutual model stress testing

OpenAI and Anthropic (Anthropic) reportedly held discussions earlier this year regarding a legally binding agreement to conduct mutual stress testing on their respective large language models. The proposed plan aimed to allow competitors to identify security vulnerabilities in commercialized models without retaining sensitive data. While it remains unclear if a final agreement was reached, the concept of peer review among AI labs has gained traction. Elon Musk recently suggested that leading AI companies should allow cross-testing of models to ensure safety. Additionally, Anthropic CEO Dario Amodei has called for third-party evaluations, and the company recently partnered with Accenture (Accenture) to assess the safety and compliance of its frontier AI models.