Commentary. The argument over artificial-intelligence regulation has reached a familiar American crossroads: genuine public risks are colliding with understandable fears that government will move too quickly, regulate badly or hand the largest technology companies an advantage. Our view is that lawmakers should act—but narrowly, transparently and with rules aimed at measurable harms rather than technological ambition itself.
The pressure is rising ahead of the November 3 midterm elections. Recent reporting describes growing concern among voters and politicians about AI’s effects on workers, safety and the infrastructure demands of data centers. Reports that banks are restricting third-party AI agents after incidents involving access to customer accounts have made the issue less theoretical. These are not arguments for banning advanced systems. They are evidence that basic controls cannot remain voluntary indefinitely. Recent coverage of the regulatory push documents both the incidents and the political response.
The strongest case for regulation is straightforward. An AI system that can move money, access private accounts, make employment recommendations or operate external software can cause damage at a speed and scale ordinary consumer-protection law was not designed to handle. Companies should be required to test high-risk systems, keep usable records of consequential decisions, disclose meaningful limitations and provide a human appeal when automated decisions materially affect a person’s livelihood or liberty.
That does not mean every chatbot needs a federal license. Nor should regulators attempt to freeze a fast-moving field around today’s technical definitions. Rules should follow capability and risk: more scrutiny for systems that can autonomously act in financial, medical, legal or critical-infrastructure settings; lighter obligations for low-risk creative and productivity tools.
The counterargument deserves respect
AI developers and their supporters are right to warn that poorly designed regulation could protect incumbents. Compliance costs that a multinational can absorb may be fatal to a small research group. Vague liability standards could encourage companies to limit useful tools, while mandatory pre-release testing could become a bottleneck controlled by political appointees or established vendors.
There is also a legitimate disagreement over who should conduct safety evaluations. Bipartisan Senate discussions reportedly stalled over whether testing should be mandatory and performed by government experts or rely mainly on company-led assessments. Neither extreme is satisfactory. Exclusive self-policing invites conflicts of interest; exclusive government testing risks delay and politicization. A better model would combine independent accredited auditors, confidential access for regulators and public reporting of standardized results. Reporting on the stalled Senate talks highlights the unresolved testing dispute.
Congress should also resist proposals so sweeping that they confuse safety with control of the technology itself. Recent proposals from lawmakers have ranged from expanding developer liability to creating a new federal AI department and restricting systems deemed more capable than humans. Those ideas reflect real anxiety, but broad prohibitions invite arbitrary enforcement and could push development overseas without eliminating demand for the tools. Coverage of the competing legislative approaches captures the breadth of the debate.
A practical test for lawmakers
The question should not be whether AI is “good” or “bad.” It should be whether a rule identifies a specific harm, assigns responsibility to an actor able to prevent it and can be enforced without discriminating against smaller competitors. That test would support clear obligations for data security, incident reporting, deceptive synthetic media, unsafe autonomous actions and discrimination in high-stakes decisions. It would reject rules written mainly to satisfy headlines or preserve a favored business model.
Innovation matters, especially for medicine, education, accessibility and scientific research. But innovation without accountability transfers the cost of failure to people who never consented to become test subjects. The alternative to sensible regulation is not freedom; it is a patchwork of lawsuits, emergency bans and private terms of service that ordinary users cannot negotiate.
Our newsroom’s position is therefore deliberately limited: regulate high-risk conduct, require independent scrutiny, protect due process and revisit the rules as evidence changes. The public does not need politicians to promise that AI will solve everything, nor industry executives to insist that every safeguard is an existential threat to progress. It needs enforceable boundaries before the next preventable failure makes the choices for us.