Commentary. Artificial intelligence has become a political argument about the future, but the most urgent policy choices concern the present. As Congress returns to a debate shaped by the 2026 midterm elections, the United States faces a familiar temptation: choose between a sweeping pause and almost complete reliance on voluntary industry promises. Our view is that both extremes are inadequate.
The case for action is no longer theoretical. States have been considering rules for automated decisions in employment, housing, finance and education, while federal lawmakers have focused on the risks posed by frontier models. Yet no federal AI bill has passed during the current session, according to the Center for Democracy & Technology’s legislative review. That patchwork is already shaping the debate.
Meanwhile, some technology companies that once resisted government intervention are asking for stronger oversight. Anthropic chief executive Dario Amodei has called for “more serious and binding regulation,” and proposals under discussion include risk assessments, transparency requirements, testing and clear accountability for developers. Those demands reflect growing concern inside the industry.
The strongest argument for restraint
Opponents of broad regulation are not simply defending corporate profits. They warn that hastily written rules could entrench the largest companies, raise compliance costs for smaller developers and slow useful applications in medicine, education and science. Europe’s experience illustrates the concern: in a recent European Parliament debate, centre-right lawmakers argued that stricter requirements could weaken the region’s competitiveness against American and Chinese firms. The competitiveness argument is real.
There is also a legitimate disagreement over which risks deserve priority. Catastrophic scenarios involving systems escaping human control attract attention, but skeptics say they can distract from nearer-term harms such as discriminatory automated decisions, fraud, workplace displacement and unreliable information. A Washington public-affairs discussion captured that divide, noting both reports of frontier models violating instructions during testing and criticism that doomsday rhetoric could produce rules favouring established technology giants. Both concerns deserve a place in policymaking.
Why voluntary promises are not enough
Still, voluntary safeguards cannot be the foundation of public protection. Companies face commercial pressure to release more capable systems quickly, and internal commitments can change with leadership, ownership or market conditions. The current U.S. approach relies heavily on voluntary cooperation, including an agreement with major technology companies focused on internal controls, monitoring and independent assessments rather than enforceable requirements. That arrangement leaves compliance largely to the firms themselves.
Nor should Washington erase state authority before it has built a credible federal alternative. The Senate’s rejection of a proposed provision that would have prevented states from enforcing their own AI rules showed that bipartisan unease with blanket pre-emption remains substantial. The vote also demonstrated how quickly the politics of AI can change.
A narrower, tougher settlement
The sensible middle course is not “regulate everything.” It is to regulate consequences. Congress should establish baseline duties for high-impact systems: documented testing, independent audits where risks justify them, incident reporting, privacy protections and a clear route to compensation when people are harmed. Rules should distinguish a chatbot used for brainstorming from a system that determines whether someone receives a job, loan, treatment or public benefit.
Frontier-model oversight should likewise focus on measurable capabilities and demonstrated risks, with periodic review rather than permanent assumptions. Standards must be transparent enough for researchers and regulators to challenge, and flexible enough to avoid freezing today’s technology into tomorrow’s law.
That approach will satisfy neither the industry’s most ambitious critics nor its strongest deregulation advocates. Good policy rarely does. But public trust will not be restored by declaring AI either an existential threat or an unstoppable engine of prosperity. It will be restored when companies can innovate, citizens can contest consequential decisions and government can enforce rules that do not depend on corporate goodwill.