
Washington has become remarkably good at turning difficult questions into political identities. Before long, Americans are no longer arguing about how to solve a problem. They are arguing about which side they are on. Artificial intelligence may still have a chance to escape that fate.
Pennsylvania Sens. John Fetterman, a Democrat, and Dave McCormick, a Republican, recently offered an encouraging example. Writing jointly in The Free Press about AI regulation, they did something increasingly unusual in Washington: They began not with their disagreements, but with the problem they share. That matters almost as much as the particular answers they propose.
AI will transform the economy, national security, education, medicine and probably much else we cannot yet anticipate. The rules governing something that consequential cannot sensibly be rewritten every four years depending upon which party controls Washington. Businesses need predictability. Workers need durable protections. National security requires continuity. And a technology racing ahead of government’s ability to understand it demands policymakers willing to learn from one another rather than retreat immediately to partisan corners.
Fetterman and McCormick reject both poles of an increasingly overheated debate: predictions that AI may destroy humanity and the assumption that technological progress should proceed largely unconstrained. Their alternative is less dramatic — and considerably more useful.
America should compete aggressively in AI, particularly against China, while protecting children and workers, addressing the effects of massive data centers on surrounding communities, imposing meaningful legal responsibility when AI systems cause serious harm and closely overseeing models capable of creating biological, cyber or nuclear dangers. Most important, they acknowledge something politicians rarely enjoy acknowledging: Government does not yet know enough to regulate everything.
That uncertainty argues for concentrating first on principles capable of surviving technological change. Liability is one. We may not know precisely what artificial intelligence will be capable of five years from now. We can nevertheless establish that companies developing extraordinarily powerful systems bear responsibility when foreseeable failures cause catastrophic harm.
The voluntary AI accord recently announced by the White House points in the same direction, although tentatively. Six major technology companies agreed to internal safeguards, outside safety reviews and board-level oversight. That is useful. But voluntary commitments by companies developing enormously profitable technologies cannot be the final word. Some responsibilities ultimately require law. And durable law generally requires something else: political consensus.
That may be the larger lesson of the Fetterman–McCormick effort. AI has not yet become completely tribal. There is no inherent Republican position on artificial intelligence and corresponding Democratic one. There are competing judgments about innovation, regulation, employment, privacy, national security and risk. Those disagreements are real. They need not become partisan articles of faith. Indeed, uncertainty should encourage cooperation. Neither party possesses a monopoly on understanding a technology nobody fully understands.
There will never be one government switch capable of turning AI’s risks off. Regulation will require experimentation, adjustment and humility as the technology changes. Fetterman and McCormick have not solved that problem. They have demonstrated something more modest but perhaps equally important: how two senators from different parties can begin solving it together.
Before artificial intelligence changes nearly everything else, perhaps it can help Washington remember how governing is supposed to work.


