Commentary. The argument over artificial-intelligence regulation is often presented as a choice between innovation and control. That is a false choice. The real question is whether governments can establish rules that protect the public without freezing a fast-moving technology in place.
Recent developments show why the debate matters. The European Union’s AI Act entered a new enforcement phase in August, giving European authorities powers over general-purpose models and other covered systems. In the United States, by contrast, federal policy has leaned more heavily on existing civil and criminal law, while states continue to propose or pass their own measures. California has also moved toward requiring a verified “kill switch” for frontier models.
This patchwork is inconvenient for companies, but inconvenience is not the same as harm. A technology used in hiring, credit, education, health care and public administration should not be governed solely by voluntary promises from its suppliers. Nor should the public have to discover a system’s weaknesses through a discriminatory denial of housing, a fabricated allegation or an unsafe automated decision.
The case for restraint
Industry’s objections deserve a fair hearing. Rules written too broadly can burden small developers that lack the compliance teams of the largest technology companies. Vague definitions may invite arbitrary enforcement. Requirements that force companies to reveal sensitive technical information could expose trade secrets or create security risks. And a patchwork of state laws can make it harder to offer the same product nationwide.
Those concerns are not arguments for doing nothing. They are arguments for better legislation. Regulators should distinguish between a low-risk writing assistant and a model embedded in a benefits system, a hospital workflow or critical infrastructure. Compliance obligations should be proportionate, predictable and backed by technical expertise. Independent audits and clear appeal procedures are more useful than symbolic disclosure forms that no ordinary person can understand.
The federal government’s preference for existing law has one advantage: it avoids duplicating rules that already address fraud, discrimination, consumer protection and negligence. But that approach also leaves uncertainty about who is responsible when an autonomous system acts unpredictably. Traditional law can punish wrongdoing after the fact; it is less effective when victims cannot identify the decision-maker, preserve evidence or afford a long legal fight.
Accountability must be practical
Our view is that policymakers should adopt a narrow, enforceable baseline rather than a sweeping code aimed at every possible future use of AI. That baseline should require providers and deployers to document high-impact systems, test them before release, monitor them after deployment and notify affected people when automated tools materially shape decisions about them.
It should also require a meaningful human review process. “Human in the loop” is not enough if the reviewer merely approves a machine’s recommendation under time pressure. Reviewers need authority, relevant information and a way to override the system without penalty.
Children and emotionally vulnerable users warrant stronger protections. Recent state and federal proposals have focused on chatbot companions, parental consent and limits on personalization for minors. Such measures may be difficult to enforce perfectly, but the difficulty of protecting young users is not a reason to abandon the effort. Platforms already make complex judgments about age, advertising and content; they can be required to explain how those judgments work.
There is also an overlooked economic issue. The expansion of data centers can place new demands on electricity grids and infrastructure. Proposals to prevent operators from shifting those costs onto ordinary ratepayers deserve serious consideration. Innovation should not depend on quietly socializing its most immediate expenses.
A workable compromise
The best regulatory model will not satisfy either extreme. It will disappoint those who want government to guarantee that AI is safe before it is released, and those who want companies left entirely to market discipline. That is acceptable. Good public policy is often less dramatic than the arguments surrounding it.
Governments should coordinate on core rights and safety expectations while allowing room for national experimentation. Companies should be free to compete on models and applications, but not on whether people receive notice, recourse and protection from foreseeable harms. And lawmakers should review the rules regularly, using evidence rather than headlines to decide what works.
AI will continue to advance regardless of whether regulation is elegant. The choice is whether accountability advances with it. A regulatory vacuum is not neutrality; it is a decision to let the strongest institutions define the terms. That is not the future a democratic society should accept.