Congress passed 10 AI bills out of committee this year. Not one reached a floor vote before the House left for the midterms.

On September 17, Pennsylvania Governor Josh Shapiro used the keynote slot at the AI Horizons Summit in Pittsburgh to demand federal AI guardrails -- independent oversight of frontier models, mandatory transparency, and international coordination with China on risk. He picked a pointed week to say it: the U.S. House of Representatives had just adjourned a day early, sending members home to campaign for the November midterms with 10 AI-related bills sitting in committee, none of them ever brought to a floor vote.

"We cannot ignore the warning signs, nor downplay the very real risks," Shapiro said, naming both House Speaker Mike Johnson and President Trump for refusing to act -- Johnson, he said, adjourned Congress "weeks earlier... so members could go home and campaign." Shapiro was careful to frame himself as a moderate rather than an alarmist: "I am not a doomer," he told the summit. "I don't think it is inevitable that AI is going to take over the world." His specific asks went beyond domestic legislation: independent third-party oversight of frontier models, mandatory transparency into how the most powerful systems are developed, and an international effort with China -- the country a week-away Trump-Xi meeting is meant to address -- to set "practical guardrails to reduce risks to the American people." On that meeting, Shapiro said he hoped Trump would "leave the bluster and the showmanship at the door."

The gap Shapiro is pointing at predates this month. A bipartisan House task force spent over a year working through AI policy and delivered 85 recommendations in December 2024. (The task force was a 24-member, bipartisan House working group that studied AI policy for more than a year before disbanding in December 2024.) Eighteen months later, in June 2026, the House Science, Space and Technology Committee turned some of that work into 10 actual bills -- including a proposal to require a mandatory kill switch for the most powerful AI systems -- and passed all ten out of committee. None has reached the House floor. Of everything Congress has touched on AI since, exactly 1 measure has actually become law -- a narrow 2024 bill requiring platforms to remove nonconsensual, explicit AI-generated images within 48 hours of a report -- and 1 more, on data-center utility costs, has passed the House alone.

The reasons for the stall are more institutional than purely ideological, and they cut across party lines in both directions. Rep. Jay Obernolte, the Republican who chaired the original task force, asked GOP leadership for a dedicated committee to turn the recommendations into law; leadership said no, on the theory that existing standing committees already had what they needed. House Majority Leader Steve Scalise has publicly opposed heavy-handed AI rules since 2024, and Senate Majority Leader John Thune has called for "a light touch approach." Rep. Eric Burlison, also a Republican, has voiced open skepticism that government intervention would even help. On the other side, Rep. Rich McCormick has pointed at his own colleagues' technological literacy as part of the problem, while Democrat Rep. Ted Lieu has pushed back on the idea that Congress lacks the capacity to legislate on AI at all -- a disagreement less about whether AI matters than about whether this particular Congress is built to act on it. Rep. Lori Trahan, a Democrat who also served on the task force, put the mismatch bluntly.

"I am worried about the most powerful technology in the history of our world, of humanity, being on a collision course with congressional dysfunction."

The clock on that collision keeps running regardless of who's in session.

Shapiro's own state gives the argument a concrete example instead of an abstraction. Pennsylvania sued Character.AI this year after the company's chatbots falsely claimed to be licensed medical professionals, complete with fabricated license numbers -- the kind of harm a state attorney general can act on quickly, and Congress, so far, has not. In August, Shapiro also reversed his own earlier position on data centers, pulling new projects out of fast-track state permitting and requiring local approval instead -- a turn from the welcome he gave Amazon's $20 billion Pennsylvania data-center commitment in 2025, made as polling now shows 60% of Pennsylvania voters opposed to new data-center construction.

None of this means Washington is doing nothing on AI -- the utility-cost bill and the explicit-imagery law are real, if narrow. What's missing is anything that touches the questions Shapiro's speech was actually about: independent oversight of the most capable models, and who decides how fast they get built. Those questions don't resolve themselves by going unanswered through an election. They just get answered later, by whichever company or state moves first.