Commentary. Artificial intelligence has moved from a specialist policy question to an election issue. That shift is welcome—but Congress is still treating the subject as if time were abundant. It is not.
Recent reporting shows bipartisan momentum for federal AI oversight ahead of the November midterm elections, even as negotiations in the Senate have stalled over basic questions such as mandatory testing and the role of government evaluators. At the same time, California and New York are advancing their own safeguards, creating the patchwork that federal lawmakers have long warned about. Reporting on the congressional impasse and state action makes the central contradiction plain: politicians increasingly agree that AI needs rules, but disagree on who should write them and how enforceable they should be.
The case for action
Our view is straightforward: the United States should establish a modest federal safety floor now, rather than wait for a comprehensive law that may never arrive. That floor should require meaningful pre-release testing for the most capable systems, independent evaluation of high-risk applications, clear incident reporting and legal responsibility when companies deploy products they knew—or should have known—were unsafe.
This is not an argument for stopping technological progress. It is an argument for applying familiar principles of public accountability to a technology whose failures can scale rapidly. Financial firms, drug manufacturers and aviation companies do not receive a permanent exemption from oversight because innovation is valuable. AI companies should not either.
There is also a practical reason to act federally. Different state rules may encourage experimentation, but they can impose conflicting obligations on smaller developers and make compliance expensive. A national baseline could protect consumers while preserving room for states to adopt stronger protections. It would also give American companies clearer expectations as they compete internationally.
The strongest counterargument
The opposing case deserves more than a slogan about “innovation.” Artificial intelligence is developing quickly, and poorly designed rules could entrench the largest firms, discourage open-source research or push investment overseas. A government approval process for every new model could become slow, politicized and technically obsolete before it takes effect.
Those concerns are real. Regulation should not demand that officials predict every future use of AI, nor should it treat a small startup like a company operating a frontier model used by millions. Rules must be proportionate, risk-based and regularly revised. They should focus on demonstrable harms—fraud, discrimination, privacy violations, dangerous system behavior and critical-infrastructure risks—rather than vague promises to make algorithms “safe.”
Industry-led testing also has value. Developers understand their systems better than most regulators, and voluntary cooperation can move faster than legislation. Recent proposals have emphasized company self-policing alongside independent auditors, a model that could help bridge the gap between technical expertise and public oversight. Coverage of those proposals also illustrates their weakness: voluntary commitments depend on companies policing themselves even when commercial incentives point the other way.
A workable compromise
The answer is neither a blank cheque nor a bureaucratic freeze. Congress should require disclosure of serious incidents, establish independent testing for the highest-risk systems and authorize a specialist regulator with narrow, clearly defined powers. Smaller developers should receive scaled obligations, while companies deploying AI in hiring, healthcare, policing, education and essential services should face stronger scrutiny.
Most importantly, lawmakers should separate rules for current harms from speculative debates about superintelligence. The public does not need to wait for a science-fiction scenario to demand transparency about automated decisions that affect jobs, credit, safety or access to services.
Election-year attention can produce theater. It can also create an opening. If lawmakers reduce the debate to “innovation versus regulation,” both sides will lose—and the public will inherit the consequences. The better choice is disciplined oversight: enough to make powerful companies answerable, flexible enough to let useful technology develop, and enforceable enough to matter.