Sen. Cantwell Pitches U.S. AI Safety Playbook Built Around Federal Testing Regime

Sen. Maria Cantwell, the top Democrat on the Senate Commerce Committee, has released a set of principles that sketches what a U.S. federal AI safety regime could look like. At the heart of her proposal is a strong role for Washington in directly testing powerful AI models, instead of leaving safety checks largely to tech companies themselves.

The framework envisions federal agencies defining common safety standards and running pre‑deployment tests for “catastrophic risk” on high‑risk AI systems. It also calls for clear rules on transparency and user disclosure, guardrails against discriminatory algorithms, and stronger accountability when AI is used in sensitive areas like employment or finance. For now, these ideas are not yet a bill, and Cantwell is not formally part of the separate Republican‑led effort to draft AI legislation—highlighting how fragmented the debate in Congress remains.

The U.S. currently relies on executive orders and agency guidance rather than a comprehensive AI safety law. Cantwell’s blueprint is still only a policy marker, but if pieces of it are translated into legislation, developers of frontier models could face routine government testing and audits before launch. Globally, the proposal will be read against the backdrop of the EU’s AI Act, and could signal that Washington is inching closer to its own overarching AI rulebook.