The US Department of Commerce’s Center for AI Standards and Innovation (CAISI) announced that Google, Microsoft, and xAI have agreed to voluntarily submit their AI models for pre-release safety testing and capability evaluations.

The new agreements expand on similar pacts reached with OpenAI and Anthropic during the Biden Administration. CAISI has now conducted 40 evaluations of AI tools, including testing of certain “state-of-the-art models that remain unreleased.” The center did not specify which models were blocked from public release.

Expanding Oversight

“These expanded industry collaborations help us scale our work in the public interest at a critical moment,” CAISI director Chris Fall said.

The cooperation marks a subtle shift in the Trump administration’s AI policy. Despite President Trump signing executive orders last year centered on “removing red tape” around AI development, the White House appears to be adjusting its stance as AI expands into military applications and Anthropic claims its Mythos model is “too powerful for public release.”

Companies Under Scrutiny

Notably, xAI’s Grok chatbot has faced widespread public scrutiny over image processing controversies, including incidents where the tool undressed people in images. Google’s Gemini model is now being used by US defense and military agencies. Microsoft’s CoPilot remains a leading enterprise AI product.

Senior Trump administration officials also met with Anthropic CEO Dario Amodei last month, even as the company is embroiled in a lawsuit with the US Department of Defense over its refusal to lower safety guardrails for government use of its models.

Representatives of Google, Microsoft, and SpaceX (which now controls xAI) did not respond to requests for comment.