We the People of these United States have many epochal achievements to our credit, but when the humans of 2500 look back at our era, Americans’ pioneering achievements in artificial intelligence may loom largest of all. Then again, maybe there won’t be any humans by then, precisely because AI has long ago wiped us all out. Perhaps whatever intelligences survive will look back on us with fondness as well as pity. These are heady times. It has suddenly dawned on the broad American public that what we do about AI right now has world-historical importance.
Last week, Speaker of the House Mike Johnson explained that this issue’s outsized importance is a good reason to convene industry leaders at the White House and hash out some agreement to mitigate extreme risks. He did not think much of calls for Congress to take up the issue without delay:
“Here’s the problem. If I called an emergency session today, and I brought everybody in the House and said, ‘Let’s regulate AI,’ nobody in the room knows yet what that should look like. Why? Because we don’t have the latest information.”
“We are not in the labs with the frontier labs who have taken this technology many steps further than even when our own report was published [in 2024].”
In response to a reporter’s question about ensuring that America keeps its edge over China, he answered:
“Well, that’s right. That’s exactly the point. If Bernie Sanders is in charge of this thing, he’ll just shut it all down and we’re toast, all right? Because then Xi and China overlap us and become the great superpower on the planet.”
“We’re not doing that. We’re not going to take stupid, knee-jerk reaction prescriptions on this. We’re going to do this right, wisely, carefully, to make sure that we maintain our edge.”
These are far from crazy answers. It is upsetting to imagine our legislators giving themselves over to the catastrophizing spirit now upon us and passing legislation that would cripple our world-beating AI firms. If Congress gets into the AI regulation business, a lot can go wrong. The AI firms and their allies know that. The White House knows that. American voters, only 16 percent of whom approve of Congress, surely know that, too.
Yet Johnson’s caution nevertheless deserves criticism, both because the alternatives to congressional action are much less adequate than he suggests and because Congress’ own readiness and constructive potential look better than he is willing to admit. We should absolutely want our legislature to be the place we take up the most pressing and most challenging issues, so that we can find a path forward that is broadly acceptable to all Americans. If we (and our representatives themselves) are out of the habit of thinking of Congress as up to the task, that is to our detriment.
Consider who is empowered if our legislators simply sit on their hands until they possess “the latest information” from the frontier labs. First, the tort system. Some, like the Manhattan Institute’s Judge Glock, argue that the “old-fashioned legal system” is up to the job of disciplining AI firms for their misconduct—or their agents’ “misconduct,” which they take to be the same thing. I am not so sure. There will be real difficulties in parsing who is responsible for many harms, an AI firm or a user employing its tools. Juries could show real sagacity in making sense of that confusion. Then again, faced with extremely technical information and companies easy to caricature, they might seriously mess things up. As Glock himself admits, it would be very helpful if lawmakers could clarify how tort laws apply to these novel cases.
Who else is empowered by congressional inaction? The Trump White House, whose record in dealing with frontier AI companies is decidedly mixed. On the one hand, the jury-rigged de facto preclearance system it has set up in the past few months seems like a reasonable response to dizzyingly fast innovation. On the other hand, the administration’s choice to designate Anthropic a supply chain risk when the company angered the armed services—a designation that a judge recently ruled illegal—suggests the limitations of a system in the hands of all-too-human executive branch officials. As Tyler Cowen recently argued in The Free Press, the White House’s enormous discretion could readily facilitate a dramatic, poorly justified turn against the industry. While the president has been adamant in his opposition to a proposed slowdown in AI research, some factions in his circle favor such a turn. If this administration has taught us anything, it is that legislators are not the only ones who can make rules while in the heat of momentary passions.
Turning to Congress itself, the picture is less shabby than one might imagine. As Johnson alluded to, a House Bipartisan Task Force on AI released an impressive report in 2024, as did a Senate counterpart. Legislators (and their staffs) have recognized the importance of these issues, and many have done a fair job educating themselves. In Section 1535 of the National Defense Authorization Act for 2026, Congress also set up an AI Futures Steering Committee within the Department of Defense, bringing together top military leaders to formulate proactive policies on a number of fronts. This is hardly earthshaking, but it is sensible enough.
Legislatively, there are several perfectly good irons in the fire. The bipartisan NO FAKES Act, reported out of the Senate Judiciary Committee, gives immediate hope of addressing AI-generated deepfakes of real people, a serious issue if not an existential risk. Sen. Chuck Grassley and five other senators from both parties introduced the AI Whistleblower Protection Act, which seeks to ensure that AI firms’ nondisclosure agreements with their employees do not prevent timely warnings of serious harms.
Most notably, catastrophic risks are addressed by the FRONTIER Act, co-sponsored by Democratic Rep. Lori Trahan and Republican Rep. Jay Obernolte, which would create a government-certified network of independent verifier organizations (IVOs) analogous to auditors, to be overseen by a government regulator housed in the Commerce Department. Firms investing the most in cutting-edge models, which are naturally the most threatening, would be required to provide deep access to the IVOs, which would in turn be responsible for certifying safety procedures and risk reporting. It is not entirely clear whether the bill gives enough teeth to the openness requirements; there are hefty penalties for outright obstructionism or lying, but legitimate worries about hard-to-penetrate obfuscation or underreporting remain. It’s also unclear whether a regulation targeting only internal development at leading firms is good enough in a world when some of the most threatening activity may come from people’s modifications of open-weight models.
All the same, the FRONTIER Act is a serious bill that has already undergone a serious round of public criticism and revision after a draft was released this summer. Johnson’s worries about Congress getting things wrong notwithstanding, it would be entirely reasonable to put this bill on the floor and conduct an honest-to-goodness, outcome-not-predetermined debate about its merits. Such a debate would give some members a chance to thunder at the AI companies and their undue influence, but most members would take their responsibility seriously. Legislators’ views on these topics have not yet (entirely) been submerged into partisan platitudes, and debates on the House and Senate floor would be instructive for the American people, certainly far more than social media pile-ons or invisible backroom White House negotiations. Debates around the country about data centers provide meaningful battlegrounds, but these particularized fights won’t yield any systemic solutions.
Unfortunately, as much as we now feel a sense of urgency, there is no clear deadline. We’re not going to get any action before the November 3 elections, and the lame-duck session will be dominated by efforts to keep the government’s lights on. Chances are that it will be left to the 120th Congress to act on AI. Perhaps the unusually large group of freshman representatives who will begin their tenure in 2027 will demand that the institution they have joined live up to its constitutional obligations.
Philip Wallach is a senior fellow at the American Enterprise Institute and the author of Why Congress (Oxford University Press). This essay previously appeared in The Dispatch.
Stay in the know about our news and events.