White House may review models
- Trump officials are weighing an executive order that could create an AI working group and a pre-release review process for powerful models. - Anthropic, Google, and OpenAI were briefed last week, and by Tuesday CAISI said five companies now allow early government model access. - That would mark a sharp turn from Trump’s earlier deregulatory posture and could slow or reshape frontier model launches.
The fight here is over frontier AI — the biggest, most capable models, not every chatbot update. The White House is now weighing whether the federal government should get a look at some of those systems before the public does. That is a real shift. Trump started this term talking like AI should move fast and avoid heavy rules, but the administration is now discussing a review process that looks a lot more hands-on. ### What changed this week? The immediate news is twofold. First, reporting on May 4 said the administration is discussing an executive order that would create an AI working group and examine formal pre-release reviews for new models. Second, on May 5, the Commerce Department’s Center for AI Standards and Innovation — CAISI — said Google would give CAISI access to models for evaluation before public release. ### Who would be in the room? The companies named in the White House discussions were Anthropic, Google, and OpenAI. The broader CAISI announcement now covers five firms: Google, Microsoft, xAI, OpenAI, and Anthropic. That matters because it suggests the government does not need to build a brand-new pipeline from scratch — part of the machinery for pre-release testing already exists. ### What does “review” actually mean? Basically, not a full licensing regime — at least not yet. The idea under discussion is closer to a government check on capabilities and risks before a model goes wide. CAISI says its agreements let the government evaluate models before they are in production, including on unreleased state-of-the-art models. ### Why is the White House even considering this? A lot of this seems to trace back to Anthropic’s Mythos model. Recent reporting tied internal administration discussions to concern that very capable systems could sharply raise cyber risk if released too openly or too quickly. The White House wasn't just talking abstractly about AI safety — they were reacting to a specific model. ### Is this
