Five labs. One federal program. That’s where US AI governance stands now.
NIST’s Center for AI Standards and Innovation, CASI, has formalized pre-deployment and post-deployment testing agreements with Google DeepMind, Microsoft, and XAI. Those three additions complete the picture. OpenAI and Anthropic had previously entered equivalent agreements, bringing the total to five frontier labs covered under the program. According to Cybersecurity Dive, CASI will evaluate frontier models from all three new participants before their public release.
What the agreements cover
Per Politico’s reporting, the Commerce Department’s Center for AI Standards and Innovation will conduct safety testing on new AI systems before public release, evaluating for security risks, autonomous behaviors, and potential for real-world misuse. This is pre-deployment review with federal oversight, not a voluntary checklist. Whether the agreements are legally binding, enforceable beyond reputational commitment, hasn’t been publicly confirmed by any source. The structural reality is simpler: every major US frontier AI lab has now formalized a testing relationship with the US government.
Unanswered Questions
- Are CASI agreements legally binding or reputational commitments?
- What happens when a lab and federal reviewers disagree at pre-deployment review?
- Does the White House executive order on mandatory review move forward, and when?
The XAI angle in context
XAI’s participation is notable given its founder’s previous public criticism of federal AI oversight. According to Politico, XAI is participating despite those prior positions. Don’t expect this to be the main story. The significance isn’t which lab joined last, it’s that the roster is now complete.
The voluntary-to-formal shift
The CASI agreements mark a structural formalization of what were previously voluntary commitments. Whether that constitutes a “binding” shift in the legal sense depends on questions about enforceability that no source has yet confirmed. What’s clear is that the architecture of informal voluntary pledges, the dominant US AI governance model since 2023, has been replaced by a structured federal vetting process covering every major domestic frontier lab. Per CIO.com, reports also suggest the White House is separately weighing an executive order to create mandatory pre-release review requirements, which would give the CASI agreements statutory backing, though that development is reported, not confirmed.
What to Watch
What to watch
The legal enforceability question matters more than the roster. If the White House executive order on mandatory model review moves forward, the CASI agreements shift from formal-but-advisory to formal-and-required. That’s the trigger for compliance teams to reassess. Until then, the practical implication is reputational and structural: labs operating outside a CASI agreement would now be the outliers.
The real question isn’t whether five-lab coverage is meaningful, it is. The question is what happens next time a frontier model raises safety concerns at the pre-deployment review stage. CASI’s architecture doesn’t yet say what happens when a lab and the federal reviewers disagree. That gap is where the voluntary-to-binding distinction will actually be tested.