The US framework has a hard limit written into it: no mandatory licensing
The voluntary framework reviewed with OpenAI, Anthropic, Google and Meta allows developers to give the government early access to certain frontier models for up to thirty days before release. It explicitly cannot be used to create a mandatory licensing or preclearance system. That constraint is the most consequential sentence in the whole arrangement.
Thirty days of pre-release access is real inspection authority, and more than any prior American instrument in this area has claimed. The no-licensing clause is what keeps it from becoming an approval gate — the government may look, and may not withhold permission to ship.
That is a coherent design and it is also where the leverage disappears. A reviewer who can examine but not act depends entirely on the developer choosing to respond to what the review finds. There is no published mechanism describing what happens if a thirty-day look surfaces something serious.
The details of what the tests measure and whether results are ever published remain undisclosed. So the regime is voluntary, non-binding, and non-public — three properties that individually have defences and together leave nothing an outside party can verify. The contrast with Brussels is now a matter of record rather than rhetoric.
Bloomberg — OpenAI, Anthropic, Google to join White House AI safety meeting → · CNBC — White House to host AI companies Tuesday to review new model-testing framework → · Indianapolis Business Journal — OpenAI, Anthropic, Google to join White House AI safety meeting →