US Government Collaborates with AI Firms for Pre-Release Model Reviews
The US government has established agreements with Google DeepMind and xAI to conduct reviews of early versions of their new AI models prior to public release.
The Center for AI Standards and Innovation (CAISI), a division of the US Department of Commerce, announced the agreements on Tuesday. The agency emphasized that the review process is vital for understanding the capabilities of emerging powerful AI models and for safeguarding US national security. According to CAISI, these collaborations will enable the federal government to "scale (its) work in the public interest at a critical moment."
"Independent, rigorous measurement science is essential to understanding frontier AI and its national security implications," said Chris Fall, CAISI director.
CAISI serves as an agency designed to facilitate cooperation between the technology industry and the federal government in developing standards and assessing risks related to commercial AI systems. The agreements with the AI companies primarily focus on identifying national security risks associated with cybersecurity, biosecurity, and chemical weapons.
OpenAI and Anthropic entered into similar agreements with the Biden administration two years ago. CAISI noted that it has already completed more than 40 such evaluations, including assessments of unreleased AI models. It is common practice for developers to share unreleased AI models with the government that have reduced or removed safety guardrails, CAISI stated in its press release. This approach assists the government in "thoroughly evaluating national security-related capabilities and risks," the agency added.
The new agreements emerge amid increasing concerns that the latest and most powerful AI models, such as those developed by these firms, could pose dangers if released publicly. AI safety experts, government officials, and technology companies worry that the extensive capabilities of these models might enable hackers to exploit cybersecurity vulnerabilities on an unprecedented scale. Anthropic has limited the rollout of its Mythos model to a select group of companies and initiated a collaborative effort to unite tech companies "to secure the world’s most critical software."
On Monday, reports indicated that the Trump administration was considering a potential executive order to establish a government oversight process for these AI tools; however, the administration has described this reporting as "speculation." Google and xAI did not immediately respond to requests for comment.
Microsoft announced a comparable agreement on Tuesday in the United Kingdom with the government-backed AI Security Institute, which also concentrates on safe AI development.
"While Microsoft regularly undertakes many types of AI testing on its own, testing for national security and large-scale public safety risks necessarily must be a collaborative endeavor with governments," Microsoft stated in a blog post regarding the two agreements.
