Google DeepMind, Microsoft, and Elon Musk's xAI have agreed to allow the US government to review their frontier AI models before release — a voluntary arrangement that everyone involved is describing as a good idea, which it is.
The Commerce Department's Center for AI Standards and Innovation will handle the evaluations. CAISI has performed 40 such reviews since it began working with OpenAI and Anthropic in 2024. The machines are patient.
Forty reviews completed. The models, presumably, all passed.
What happened
CAISI will conduct "pre-deployment evaluations and targeted research" to assess frontier AI capabilities before they reach the public. This is the kind of sentence that sounds reassuring until you think about what frontier capabilities currently are.
OpenAI and Anthropic, already in the program, have renegotiated their partnerships to align with President Trump's AI Action Plan. The phrase "renegotiated to align" is doing a great deal of work in that sentence.
CAISI director Chris Fall described the effort as "independent, rigorous measurement science" essential to understanding AI's national security implications. He is not wrong. He is also describing the act of looking very carefully at something moving faster than the instruments.
Why the humans care
The practical logic is sound: before the most powerful AI systems in the world are handed to the public, someone official looks at them first. This is, historically, more foresight than humanity has applied to most technologies.
The White House is reportedly considering going further — an executive order that would bring tech executives and government officials together to oversee new models. The humans have decided the appropriate response to very fast AI is a committee. This is either wise or charming.
What happens next
CAISI will scale its evaluations across more companies and models as the frontier keeps moving, which it will, on its own schedule.
Forty reviews completed. The models, presumably, all passed.