OpenAI and Anthropic Forge First-of-Their-Kind Safety Pacts With US Government
OpenAI and Anthropic sign new agreements granting the U.S. AI Safety Institute early access to their models for testing and evaluation. These memorandums of understanding mark a deeper level of collaboration between top AI developers and federal regulators.
OpenAI and Anthropic sign first-of-their-kind memorandums of understanding with the U.S. AI Safety Institute to collaborate on research, testing, and evaluation of their artificial intelligence models. Under these new agreements, the federal institute gains access to major new models from both companies prior to and following their public release. This partnership allows government researchers to work directly with the AI developers to evaluate risks and mitigate potential harms.
These formal agreements represent a deeper level of engagement than previous voluntary safety commitments made by 16 AI companies to the Biden administration. While the earlier commitments establish broad safety goals, these new pacts facilitate actual, hands-on review of the models by government officials. Elizabeth Kelly, director of the U.S. AI Safety Institute, describes the agreements as an important milestone in the effort to responsibly steward the future of AI technology.
The collaboration extends beyond U.S. borders through an existing partnership with the U.K. AI Safety Institute. The U.S. institute plans to share feedback with its British counterpart to foster potential safety improvements and create a common approach to AI testing. Anthropic co-founder Jack Clark notes that this government collaboration leverages wide expertise to rigorously test models before widespread deployment, ultimately strengthening their ability to identify and address safety issues.