The Center for AI Safety and Infrastructure (CAISI), operating within NIST and the Department of Commerce, has reached what amounts to full coverage of the major US frontier AI labs. Google DeepMind, Microsoft, and xAI have joined OpenAI and Anthropic in agreeing to share frontier models with the agency for pre-release evaluation. Per NIST’s CAISI page, the agency has completed more than 40 such evaluations, including assessments of state-of-the-art models that remain unreleased to the public. Evaluation scope covers cybersecurity risks, biosecurity, and chemical weapons.
That five-lab coverage is already significant. But the story moving at the policy level is what may come next.
The executive order discussion
Reuters reports that the White House is considering an executive order to establish a working group on AI that would bring together tech executives and government officials. Forbes characterizes it similarly: a working group under consideration, not a finalized mechanism. These reports confirm that discussions are active. What they don’t confirm is the specific structure the Wire’s research characterized as a “mandatory review process”, Reuters’ framing is “discussing” and “working group,” not mandatory review. The distinction matters for compliance planning: a working group is advisory, a mandatory pre-release review requirement would be a structural change to how frontier AI ships in the United States.
The White House has not confirmed the EO’s existence or scope. Treat the current state as: discussions confirmed, mechanism unconfirmed, mandatory framing unconfirmed.
Why it matters
The voluntary-to-mandatory question is the central tension in US AI governance right now. CAISI’s agreements work because the major labs have chosen to participate. That’s not nothing, five labs, 40-plus evaluations, including unreleased models, is a functioning review program. But voluntary frameworks carry an inherent constraint: they bind only willing participants, and only for as long as participation remains in those participants’ interest.
An executive order establishing a formal structure, even a working group, would be the first step toward a durable institutional home for pre-release review. Whether that home eventually hosts a mandatory review requirement, or stays advisory, is the question the hub will be tracking through whatever comes next out of Washington. The non-obvious implication worth flagging: if an EO establishes a working group that then recommends mandatory review, the compliance obligation timeline for frontier labs could compress faster than a standard regulatory notice-and-comment process would suggest, working groups can move to recommendations in months, not years.
Context and precedent
CAISI’s expansion to five labs didn’t happen in a vacuum. The hub’s earlier reporting on Anthropic’s restricted “Mythos” model, which triggered White House meetings on kill-switch design and questions about who governs AI systems too capable for public release, established the policy context in which this EO discussion is reportedly happening. The Reuters and Forbes reports don’t name Mythos as a specific catalyst; treat any direct causal link between the model and the EO as earlier reporting context, not confirmed in this brief’s sources.
The US approach also sits in direct contrast to the EU’s August 2026 deadline for high-risk AI compliance under the EU AI Act, a mandatory framework with defined penalties. Washington is still working from voluntary agreements and working group proposals. That gap is the regulatory divergence the hub has flagged consistently across 2026 coverage.
What to watch
Watch for a formal White House announcement or executive order publication. If an EO issues, the critical details are: whether it creates a mandatory review requirement or an advisory working group, which agency hosts it, whether CAISI’s existing agreement structure is folded in or parallel to it, and what the timeline and scope look like. A working group recommendation is a slower path to mandatory review than a direct EO mandate; both are worth tracking, but they have different compliance implications.
TJS synthesis
CAISI’s five-lab coverage is the stable, confirmed development here. The EO discussion is real but unsettled, Reuters says “considering,” Forbes says “considering,” and the White House hasn’t confirmed. For compliance planning purposes: the voluntary framework is functioning now, and any mandatory mechanism is at minimum months away from having enforceable requirements. The productive question isn’t “will this become mandatory?”, it’s “if it does, are our evaluation and documentation practices already consistent with what CAISI’s scope suggests a review program would assess?” Cybersecurity, biosecurity, chemical weapons. That’s the current evaluation framework. Start there.