Google, Microsoft and xAI accept U.S. AI model safety checks

Google, Microsoft and xAI have agreed to share unreleased AI models with the U.S. Department of Commerce for safety testing before those systems are released to the public, according to a statement from the Commerce Department.

The testing will be handled by the Center for AI Standards and Innovation, or CAISI, a Commerce Department office focused on AI evaluation. CAISI said it will examine the models for national security risks, with particular attention to cybersecurity, biosecurity and chemical weapons.

Chris Fall, CAISI’s director, said in a statement that “independent, rigorous measurement science is essential to understanding frontier AI and its national security implications.” He said the expanded industry collaborations will help the agency scale its work “in the public interest at a critical moment.”

The agreement adds Google, Microsoft and xAI to a group of companies that have already allowed CAISI to inspect their models. OpenAI and Anthropic were the first AI developers to work with the agency after reaching an agreement with the Biden administration about two years ago, according to the article.

CAISI has already completed dozens of evaluations, including tests on unreleased models, the article said. That suggests the agency has been building a process for reviewing systems before they are widely deployed, even as AI companies continue to release more powerful models.

The new arrangements come as the Trump administration appears to be weighing a more formal role for the federal government in AI oversight. The New York Times reported earlier this week that the administration is discussing an executive order that would create a working group of tech executives and government officials to design a formal review process for new AI models.

That would mark a shift from the White House’s earlier hands-off posture. The article said the current administration has generally taken a laissez-faire approach to AI development as the U.S. tries to keep pace with China.

Public concern around AI has also widened beyond model capabilities. The article cites fears around job loss, mental health and cybersecurity, alongside questions about what happens when advanced systems are released without enough external testing.

Trump said last year at an AI event that the government should not try to stop AI through politics or “foolish rules and even stupid rules,” according to the article. The new talks suggest the administration may be moving toward a more careful framework while still trying to avoid broad regulation.

The article also notes that CAISI’s work is not new. It has already carried out multiple model evaluations, including on unreleased advanced systems, and the addition of Google, Microsoft and xAI expands that testing program further.