Trump's AI testing framework sidelines open-source and sets no binding rules

Three days after the EU AI Act became enforceable, Washington published a voluntary AI evaluation framework that excludes open-source models. The transatlantic regulatory gap just became official.

In plain terms: The US government published AI safety testing guidelines. They're voluntary, exclude open-source models, and name no enforcement authority - a direct contrast with EU rules that took effect on August 2nd.

The fact

The White House released an AI testing framework that applies to closed commercial models and explicitly carves out open-source systems, according to The Verge. The document sets no binding timelines, no penalties, and designates no federal agency as enforcer. It arrives three days after the EU AI Act's GPAI rules became legally binding across 27 member states.

Our take

The contrast is stark. The EU framework compels frontier model providers to document capability evaluations, report serious incidents, and cooperate with the European AI Office. Washington's response is a voluntary checklist that doesn't reach the models most widely deployed at scale—open-weight systems from Meta, Mistral, and others. This isn't regulatory caution; it's regulatory absence. The practical effect: US-based AI infrastructure sectors—healthcare, finance, critical infrastructure—operate with no federal floor on AI testing requirements while EU competitors face mandatory audit obligations.

What to watch

Whether any Congressional committee converts this into binding legislation, and how the EU AI Office treats US companies that comply with the American framework but not the EU one.