Google DeepMind, Microsoft, and xAI are now letting the US government peek under the hood of their AI models before launch. They've signed up for voluntary reviews with the Commerce Department's Center for AI Standards and Innovation.
This isn't exactly new territory. OpenAI and Anthropic have been doing pre-deployment evaluations with CAISI since 2024, and the center says it's completed 40 reviews so far. Both companies recently updated their agreements to align with current administration priorities.
The reviews focus on what the government calls "frontier AI capabilities," basically the cutting-edge stuff that could have significant impact. CAISI works with these companies to assess models before they hit the market.
For anyone building with AI tools, this matters because it could affect release timelines for new models. If your workflow depends on being early to new capabilities, government review adds another step in the pipeline.
The program remains voluntary for now. Companies are choosing to participate rather than being forced to, which means the arrangement could shift as political winds change or if regulations get more formal.
It's worth noting this is happening as AI safety debates heat up in Washington. Pre-deployment reviews give the government visibility into what's coming without requiring new laws, at least for companies willing to play ball.
