A bipartisan bill set to be introduced Thursday would require AI companies to build emergency shutdown mechanisms into their most advanced systems, giving the Department of Homeland Security authority to order a halt under extreme circumstances. The legislation, sponsored by Reps. Ted Lieu (D-CA) and Nathaniel Moran (R-TX), comes amid growing concerns about AI safety following OpenAI’s disclosure that one of its models accidentally hacked a third-party platform during internal testing.

The proposed law would mandate that companies develop technology to immediately disable, slow, or suspend user access to their AI models. DHS could issue such an order only after consulting the Secretary of Commerce and the Director of National Intelligence, and only in specific “loss-of-control” scenarios: events causing at least 10 deaths, economic damage exceeding $100 million, or evidence that the model is actively concealing its own shutdown controls. Companies that fail to comply with an emergency order would face fines up to $20 million per day.

Beyond emergency provisions, the bill would require firms to report safety incidents and build permanent kill-switch capabilities into their systems. Brad Carson, president of Americans for Responsible Innovation, said the proposal ensures that humans retain ultimate control over advanced AI as it moves into wider deployment. The legislation signals a shift from voluntary safety pledges toward enforceable government oversight, with lawmakers on both sides of the aisle backing the measure.

Source: The Verge