Trump's secret frontier AI review framework faces five questions

A new op-ed in Tech Policy Press picks at a piece of AI policy that has been running quietly since June, and does it by asking, not demanding. Michelle De Mooy, an AI governance consultant, lays out five questions the US government should answer about a framework that lets it look at frontier AI models before they ship.

The framework comes from Executive Order 14409, 'Promoting Artificial Intelligence Innovation and Security,' which President Trump signed in June. It directs a National Security Agency-led group to develop a process under which developers give the government access to their models for up to thirty days before release. The White House, per the piece, now says it will keep that framework a secret.

That secrecy is the thing De Mooy is pushing on. Her five questions are straightforward: what makes a model covered, how long can review last, who gets on the approved-access list, what recourse does a developer have, and what will the public be told. She notes that roughly 100 organizations already have access with no published eligibility criteria, and that the process for the framework's own development has 'operated informally, without published criteria, defined timelines.' She also flags that the Center for AI Standards and Innovation was reportedly directed to stop issuing public reports days after the order was signed.

Why this matters if you are not a frontier developer: enforcement is already visible in the market. The op-ed cites a nineteen-day shutdown of Anthropic's frontier models via an export control order, and a two-week gated rollout of OpenAI's GPT-5.6. Access rules, clocks, and appeal rights that are not written down anywhere can still move products on and off the market.

The honest caveat is that this is one analyst's read of a process whose technical basis may legitimately need to stay classified, and she concedes some of it should be. What the piece does not give you is the government's own defense of the current design, or a full count of how many models have been reviewed under it. Her closing line is the one worth sitting with: legitimate regulation, she argues, is not defined by how much authority the government possesses, but by how transparently that authority is exercised.