Bipartisan Senate negotiators are drafting a duty-of-care bill that would give the federal government enforceable power to block release of frontier AI models deemed unsafe, according to two Senate aides and a lobbyist cited by Reuters. Majority Leader John Thune, Commerce Chairman Ted Cruz, and Senator Amy Klobuchar are among the principals, and the proposal would cover Alphabet’s Google, Anthropic, and OpenAI while preempting certain state laws on catastrophic risks.

The frontier labs are, remarkably, lobbying for it.

On Saturday, Anthropic chief executive Dario Amodei published an essay arguing that “we must slow the pace at which we improve the capabilities of AI models,” proposing embedded third-party evaluators with employee-level lab access. Sam Altman posted on X agreeing and committing OpenAI to independent evaluators. Anthropic’s public policy chief Sarah Heck went further, calling on Congress to pass “a national law requiring testing of frontier models, with the power to block the most advanced models that prove to be unsafe.”

President Trump, speaking Sunday to reporters at the Irish Open, wasn’t buying it. “We’re leading China in AI,” he said, framing the slowdown chorus as sabotage. “We can put guardrails. We can do this and that. But I think you have a lot of negative forces that are bringing it up that shouldn’t be bringing it up.”

The legibility problem here isn’t ideological, it’s operational. Earlier ad hoc federal action against Anthropic drew a rebuke from a federal judge, underscoring the absence of a statutory framework or clear appeal path. A duty-of-care regime would regularize that power and route challenges through federal courts. Businesses that treated frontier models as utility infrastructure would face a supply-side risk they can’t hedge inside the vendor: a model could go dark mid-cycle, and the restoration timeline would sit with a judge.

For small businesses that have wired AI into lead research, outreach drafting, and content production, this is a continuity question, not a policy abstraction. The same firms that adopted Anthropic’s Fable 5.1 agent-cost reduction to run customer-facing work at scale now have single-vendor dependencies inside a regime where a court can freeze that vendor.

The calendar is doing its own work. The House has one week in session before the November 3 midterm recess; the Senate has three. Whether the bill clears both chambers or slips into 2027, the shift it signals is the durable fact: voluntary safety pledges from labs are being replaced by enforceable obligations from Washington, at the labs’ invitation. The frontier is being deputized. The question small operators face isn’t whether that’s good policy. It’s which of their workflows survive a model going offline for three weeks with no warning.

Sources