US Senate weighs frontier AI duty of care safety law

The Senate duty of care proposal would be the most concrete US move yet to put hard legal brakes on the riskiest frontier AI systems. Instead of relying on voluntary pledges and shifting safety rhetoric, labs like OpenAI, Google and Anthropic would face explicit liability if they fail to guard against catastrophic misuse, from bioweapons design to cyber operations. A narrowly targeted regime that applies only to the most capable models mirrors parts of the EU AI Act, but with stronger shutdown powers and a litigation heavy enforcement model.

For the race to AGI, this is a signal that the US political system now treats advanced model safety as a national security problem rather than a niche ethics issue. If enacted, compliance would raise fixed costs and slow the tempo of frontier releases, which entrenches the largest labs that can afford safety infrastructure and legal teams while squeezing smaller entrants. At the same time, a clear federal framework could reduce regulatory uncertainty for investors and make it easier for labs to justify big bets on long term alignment research, because the rules of the game would no longer change with every new scandal.