Politico interviewed more than a dozen lawmakers and congressional staffers about regulating artificial intelligence and came back with a body that doubts its own competence. The people it spoke to question whether elected representatives understand the technology well enough to write law about it, or can get up to speed quickly enough to matter. Rep. Sam Liccardo (D-Calif.) said some of his colleagues are still trying to spell the abbreviation for AI correctly, and that Congress needs a sharp jump in baseline knowledge before it can pass anything substantive — a jump he does not expect to happen fast. The trigger for this round was Jacob Coxson, the researcher who left Anthropic and warned in a post that went viral that the companies building powerful AI models are "playing with our lives," because they understand those systems could destroy humanity.
There is now something close to consensus on Capitol Hill that AI needs more specific regulation. There is nothing close to consensus on what the rules would say, and a visible preference among many members for handing the problem to somebody else.
Sen. Ted Cruz (R-Texas), who chairs the Senate Commerce Committee, responded to Coxson's warnings by backing legislation to prevent "catastrophic risk." He pointed to a bipartisan bill being pushed by Senate Majority Leader John Thune (R-S.D.) and Sen. Amy Klobuchar (D-Minn.), which is supposed to establish more concrete rules for AI. Neither Thune nor Klobuchar would say when they plan to introduce it. Politico's inquiries to their offices went unanswered or produced a vague comment.
At the other pole, Sen. Bernie Sanders (I-Vt.) and other progressives are calling for a ban on developing AI superintelligence — a technology that may well carry existential risk, and that nobody knows how to build or whether it can be built at all.
The comic relief is a year old and comes from the executive branch. Linda McMahon, education secretary in Donald Trump's administration, repeatedly pronounced AI as "A1" — the steak sauce — while expounding at length on the importance of chatbots in education.
The A1 story is the one that circulates, and it is the least useful thing here. Vocabulary is fixable; staff can be hired, briefings can be scheduled, and plenty of members who pronounce it correctly have no clearer idea what to regulate. The genuinely telling detail is the Thune–Klobuchar bill: the Senate's most senior Republican and a senior Democrat are jointly behind the only concrete vehicle anyone can name, the Commerce chair has publicly endorsed the goal, and none of them will put a date on introduction. That is not a knowledge gap. That is a bill being used as an answer to a question rather than as a piece of legislation.
Look at what the two active positions have in common. Cruz's catastrophic risk and Sanders's superintelligence ban are both about systems that do not exist yet. Neither addresses the models currently shipping into hiring, lending, classrooms, and government procurement. It is easier to legislate against a hypothetical, because a hypothetical has no lobbyists and no deployed revenue attached to it, and because a bill about a machine nobody has built cannot be embarrassed by a constituent's complaint next quarter. The agenda is also being set from outside: the reason there is a new round of debate at all is one researcher's resignation post. That is a legislature responding to the labs' internal politics rather than to a schedule of its own.
Rep. Chip Roy (R-Texas), who is leaving Congress soon, told Politico that the government cannot simply ignore AI and hope it turns out fine, and that an approach in which the state supposedly solves the problem by banning the technology outright will not work either. Leaders, he said, will have to act deliberately and finally take responsibility for making policy. It is the most reasonable formulation in the story, and it comes from someone who will not be there to act on it.