Senate bill would let the government block frontier model releases
A bipartisan 'duty of care' draft forces national-lab testing on top models and preempts state AI laws — but faces a narrow pre-election window.
Copy markdown
A legal 'duty of care' for the biggest labs
Reuters reports Thune, Cruz, Klobuchar and Cantwell are drafting a bill that makes preventing "catastrophic risks" a legal obligation for developers of the most capable models — Google, OpenAI and Anthropic squarely in scope. It would turn today's voluntary safety pledges into enforceable law.
Washington could gate a model launch
The draft would let the government block a model judged unsafe, with the most powerful ones tested by scientists at national labs for cyberattack, bio and nuclear misuse. Companies could challenge a block in federal court, and the exact scope of that power is still being negotiated.
One federal rule replaces the state patchwork
It would preempt states from enforcing their own laws on certain AI risks, including bio- and nuclear-weapon design. For anyone shipping across state lines, that swaps a growing patchwork of state rules for a single national standard — if it becomes law.
The clock is the real obstacle
Passage is far from certain: the House sits just one week before the Nov. 3 midterms and the Senate has three, a narrow window for a bill this contested. Watch whether it moves as standalone text or rides a must-pass bill.