Commentary. The argument over artificial intelligence is often presented as a choice between innovation and regulation. That is the wrong frame. The real choice is between building public rules before powerful systems become deeply embedded in society, or allowing a handful of companies to write those rules after the fact.
The debate has sharpened as lawmakers and technology leaders confront systems that can generate software, automate decisions and potentially perform tasks once reserved for highly trained professionals. Recent proposals in the United States range from criminal and civil liability for harmful uses to far more sweeping ideas, including restrictions on systems that exceed human capabilities. The White House has also promoted an accord centered on company self-policing and independent external audits, according to the Washington Examiner.
Our view is straightforward: voluntary safeguards are useful, but they cannot be the foundation of public protection. Companies should be free to compete and experiment, yet they must meet binding standards when their products can affect employment, education, financial access, public safety or democratic discourse.
The case for moving quickly
The opposing argument deserves more than a caricature. AI development can raise productivity, assist scientific research and make specialized services cheaper and more accessible. Slowing deployment too aggressively could preserve inefficient systems, push investment to jurisdictions with weaker oversight and deny people tools that improve their work.
There is also a practical problem with regulating technology that changes rapidly. A law written around one model or one technical capability may become obsolete before it takes effect. Broad prohibitions could favor established firms that can absorb compliance costs, while smaller developers and open-source researchers lose room to innovate.
Those concerns are legitimate. A government that treats every technical advance as an emergency will eventually discourage beneficial experimentation. Regulation should therefore target demonstrable risks and require periodic review, rather than attempt to freeze the field at its current stage.
Why self-policing is insufficient
But the alternative cannot be trust alone. Technology companies have commercial incentives to release products quickly, market them aggressively and disclose failures selectively. External audits can help, but auditors need independence, access to meaningful data and authority to publish uncomfortable findings. Without those conditions, “independent” oversight risks becoming a certification exercise.
The political proposals now circulating illustrate the danger of both extremes. A sweeping ban on systems deemed smarter than humans could be impossible to define and enforce. Yet leaving liability unclear can allow developers and deployers to shift responsibility onto users after harm occurs. The sounder approach is a graduated system: tougher testing and reporting for higher-risk models, clear duties for companies deploying them, and penalties proportionate to negligence or concealment.
That framework should also protect the public from less dramatic but more immediate harms. Automated decisions can reproduce discrimination, expose private information and make it difficult for people to challenge an outcome. These are not speculative questions about a distant superintelligence. They are governance failures that existing institutions can address now.
Rules should be democratic
Public policy must also resist the idea that technical expertise alone can settle social choices. Engineers can explain what a model does; they cannot decide how much surveillance a society should tolerate, which jobs deserve protection or what level of error is acceptable in a medical or legal setting. Those decisions belong in open political debate, with affected communities represented.
That does not mean every question should be settled by politicians reacting to headlines. Independent experts, civil-society groups and workers need a formal role in testing standards and enforcement. International coordination is equally important because models, data and harms cross borders. But global cooperation should not become an excuse for waiting indefinitely.
The technology will continue to advance. The responsible response is neither panic nor complacency. It is a regulatory system that rewards useful innovation, makes powerful firms answerable for foreseeable harm and gives citizens rights when automated systems affect their lives. The companies building these tools may be moving fast. Democratic institutions should move deliberately—but they must move.