OpenAI and Anthropic to Share Models with US Government

OpenAI and Anthropic have agreed to give the U.S. government access to fresh AI models before they are released to improve their security.

The companies have signed memoranda of understanding with the US-based AI Security Institute to ensure access to the models both before and after their public release, the agency announced Thursday. The government says the move will support them work together to assess security risks and mitigate potential problems. The U.S. agency said it will provide feedback on security improvements in collaboration with its counterpart in the U.K.

Opening up access to AI models is a significant step as federal and state lawmakers consider what kinds of safeguards to put on the technology without stifling innovation. On Wednesday, California lawmakers passed the Protected and Secure Innovation for Frontier Artificial Intelligence Models Act (SB 1047), which would require AI companies in California to take certain security measures before training advanced foundational models. The measure has been met with resistance from AI companies including OpenAI and Anthropic, who warn it could hurt smaller open-source developers, although it has since undergone some changes and is still awaiting Gov. Gavin Newsom’s signature.

Elizabeth Kelly, director of the U.S. Artificial Intelligence Security Institute, said in a statement that the fresh agreements are “just the beginning, but they represent an important milestone in our efforts to responsibly steward the future of AI.”