Congress has one working week left before the November 3 midterms. After that it’s out until November 9. Two stories from Thursday suggest something started moving in exactly that week.
The Senate is negotiating a duty of care
Reuters reports, citing two Senate aides, on draft legislation that would impose a “duty of care” on AI developers: an obligation to design their products so that catastrophic risks are prevented. That’s the core of it.
The second part has more teeth. Negotiators want to give the US government the power to block the release of certain models judged unsafe. Companies could challenge that in federal court. How much power the government would actually hold is still open.
There’s also a piece that will annoy the states: the draft would stop them enforcing their own laws covering certain model risks. One aide’s example of such a risk was a bad actor using an AI system to design nuclear or biological weapons.
It would apply to models with the most advanced capabilities. On the US side that means Google, Anthropic and OpenAI. Everything is still being negotiated.
Four members want the recess cancelled
In parallel, a letter to Speaker Mike Johnson is circulating in the House. Sam Liccardo, George Whitesides, Lori Trahan and Ted Lieu want the chamber brought back immediately and kept in session until Congress advances bipartisan AI safeguards. The two September working weeks were cancelled earlier this month so members could campaign in their districts.
The tone: “Reasonable minds may disagree about precisely how Congress should regulate this rapidly evolving technology. We cannot disagree about the imperative for Congress to act.” And, rather more pointedly: while AI safety experts and Americans increasingly urge action, Congress fiddles.
The evidence comes from Anthropic
What the letter leans on is the interesting part. Its evidence for urgency is Anthropic’s threat report from this week — the one where the company describes disrupting efforts to use Claude in ways that could support biological weapons development. Plus a Chinese government-linked surveillance and recruitment operation targeting Uyghurs in Syria, and an Iran-linked attempt to use Claude to analyse publicly accessible data and produce targeting recommendations against US naval forces in the region.
Axios adds that members of both parties floated this kind of move this week, after current and former Anthropic researchers went public with warnings that AI could wipe out humanity within the decade.
My take
Anthropic has spent years arguing that someone needs to write rules. Now its own threat report is being quoted in a letter to the Speaker, and the Senate has a draft that would land squarely on its own models. That isn’t irony, it’s the logical end of the argument — but it is the moment where you find out how much of it was meant.
Whether anything comes of it is another question. One working week before the midterms, a draft where everything is still under negotiation, and a preemption clause that takes authority away from the states: that does not look like a law by November. It looks like a text waiting on the desk of the next Congress.
Sources: Reuters: US Senate negotiators consider requiring AI firms to mitigate known major risks, CNN: House Democrats urge Speaker Johnson to cancel recess to pass AI safeguards, Axios: Mike Johnson urged to cancel House recess over AI warnings