A bipartisan effort to regulate the most powerful AI systems in the US has hit a wall. Senate Majority Leader John Thune’s push to create a national framework for AI safety has been postponed until after the Senate recess, caught between an AI company that wants stricter rules and a ranking committee member who thinks the whole enforcement approach is wrong.

What the bill actually does

The proposed legislation, backed by Thune, Senate Commerce Committee Chair Ted Cruz, and Senator Amy Klobuchar, would create a legal duty for AI developers to manage risks tied to advanced systems. Think cybersecurity threats, biological risks, and whatever else frontier AI models might cook up as they get more capable.

Two key provisions stand out. First, AI developers would be required to notify the Commerce Department whenever they achieve significant leaps in AI capabilities. Second, the government would gain authority to seek court injunctions against companies that don’t comply with their agreements.

The bill also aims to create a national standard that would preempt the patchwork of state-level AI regulations currently emerging across the country.

Why Anthropic is playing hardball

Anthropic’s opposition is genuinely unusual. The company isn’t fighting regulation. It’s arguing the proposed rules aren’t tough enough. Specifically, Anthropic wants mandatory public risk disclosures, viewing the current proposal as too lenient in how it handles transparency around dangerous AI capabilities.

The company’s position also partially aligns with some state-level regulatory approaches, which creates an awkward dynamic. The bill is supposed to preempt state rules, but one of the major AI players is saying those state rules might actually have the right idea.

The Cantwell problem

The other source of friction comes from Senator Maria Cantwell, the ranking Democrat on the Commerce Committee. Rather than the injunction model proposed in the bill, Cantwell prefers relying on expert input from the Commerce Department and national labs to test and evaluate AI technologies.

The Senate Commerce Committee markup on the bill has been deferred entirely until after the current recess.