U.S. Commerce Department Expands Pre-Release AI Testing to Google DeepMind, Microsoft and xAI

The Center for AI Standards and Innovation (CAISI) at the U.S. Commerce Department announced on May 5 that it has signed agreements with Google DeepMind, Microsoft and xAI to evaluate frontier artificial intelligence models before they are released to the public. The new agreements expand a government testing program first launched in August 2024 by the U.S. AI Safety Institute, the entity later renamed CAISI, when it established similar partnerships with OpenAI and Anthropic.

The Details

The agreements give CAISI access to frontier AI models from the three additional developers for evaluation prior to public deployment. CAISI said the expanded partnerships enable both pre-release and post-deployment assessments, including testing in classified environments and evaluations of models with reduced or removed safeguards.

The Guardian reported that the evaluations are focused on national-security-related risks, with explicit attention to cybersecurity, biosecurity and chemical weapons. CAISI says it has already completed more than 40 evaluations of frontier AI systems, including on unreleased models, before adding these three companies to its testing program.

Chris Fall, director of CAISI, said independent, rigorous measurement science is essential to understanding frontier AI and its national security implications. He also said the expanded industry collaborations help the institute scale its work in the public interest at what he described as a critical moment.

The new agreements build directly on the government testing partnerships first established with OpenAI and Anthropic in August 2024. CAISI says those earlier agreements were subsequently renegotiated to align with directives issued under Commerce Secretary Howard Lutnick and with America's AI Action Plan.

Context

The announcement represents a notable continuity in federal AI oversight policy. Although the Trump administration has generally emphasized a lighter regulatory touch on artificial intelligence, it is preserving and expanding a government-access testing channel first established under the Biden administration.

With Google DeepMind, Microsoft and xAI now included, CAISI holds voluntary pre-release testing agreements with five of the most prominent frontier AI developers operating in the United States. The institute has not publicly disclosed which specific models are queued for evaluation under the new agreements, nor has it published a schedule for when those assessments will occur.

Separately, CNBC independently reported that the White House has discussed creating an AI working group that could explore broader oversight procedures for vetting models before public release. However, CNBC also cited a White House statement saying discussion of potential executive orders is speculation and that any policy announcement would come directly from President Trump. The Guardian and Forbes also covered the CAISI agreements.

What's Next

With agreements now covering Google DeepMind, Microsoft, xAI, OpenAI and Anthropic, CAISI has established a broad voluntary channel for government review of the most capable AI systems before they reach the public. The institute says it intends to continue scaling evaluations as new models reach the threshold for frontier capabilities.

The prospect of a broader White House oversight process or executive order remains unconfirmed. The White House statement characterized such reports as speculative, meaning any formal expansion of pre-release vetting beyond the existing voluntary CAISI agreements would require an explicit announcement from the administration. For now, the pre-release testing framework remains a voluntary, institute-led initiative rather than a mandatory regulatory requirement.