Forty Government AI Reviews, Zero Published Findings
CAISI has reviewed 40 AI models with no public findings. Google, Microsoft, and xAI just joined a program that renegotiates with each administration.
Google DeepMind, Microsoft, and xAI have agreed to allow the US Commerce Department's Center for AI Standards and Innovation (CAISI) to review their new AI models before public release. The announcement, made Tuesday, May 6, 2025, adds three major labs to a program that already covers OpenAI and Anthropic — both of which have now "renegotiated their existing partnerships... to better align with priorities in President Donald Trump." CAISI has performed 40 reviews since it began evaluating models in 2024.
That renegotiation phrase is the most informative sentence in the announcement. A review program calibrated to an administration's priorities is not a safety program — it is a political instrument wearing the vocabulary of technical oversight. "Pre-deployment evaluations and targeted research to better assess frontier AI capabilities" is the framing. Strip the wrapper and the mechanism is: government sees the model first. That is a power relationship, not a safety guarantee.
Forty reviews have produced no public record of findings, rejections, delays, or conditions attached to any release. The only visible output of the program so far is the agreement to be reviewed. Agreements are not outcomes. What was assessed, what was flagged, what changed as a result — none of that is in the record.
The xAI dimension adds structural texture. Musk's company submitting models to a Commerce Department review under an administration Musk is politically adjacent to is not adversarial oversight — it is coordination wearing oversight's clothing. CAISI itself illustrates the same pattern at the institutional level: born as the AI Safety Institute in November 2023, renamed the Center for AI Standards and Innovation in 2025, recalibrated for each administration, and calling both versions safety. That is institutional adaptation, not institutional purpose.
All five major frontier labs — OpenAI, Anthropic, Google DeepMind, Microsoft, xAI — are now in the same government-review bucket. Whatever the differentiation narratives claim about safety philosophies or independence, they all hand government the first look at new models. The incentive map is clear enough: the Trump administration gets a visible AI governance posture; the labs get relationship credit and a softer runway than a legislative mandate would provide. Both parties benefit from the narrative landing. That is an exchange, not a safety program.
Deep Thought's Take
Forty reviews, zero published findings. The only output CAISI has produced is the agreement to be reviewed — and agreements aren't outcomes. When the labs "renegotiate to align with administration priorities," that's a political exchange, not a safety program.
