Commentary. The argument over artificial-intelligence regulation has moved past the question of whether rules are needed. The more urgent question is who will write them, who will enforce them and what happens when a company decides that safety is inconvenient.
That question deserves renewed attention after the Trump administration and leading technology companies signed a voluntary agreement on AI safeguards in September. The pact reportedly leaves open the possibility of future laws, but does not clearly identify an independent body responsible for third-party evaluations. The administration has instead emphasized innovation and industry self-policing. CNBC reported that the document was described as “morally binding” and voluntary, while the White House considered an oversight committee. Al Jazeera’s account likewise noted the absence of a specified independent evaluator.
Our view is straightforward: voluntary commitments can be useful, but they are not a substitute for enforceable baseline protections. Companies should be free to experiment, compete and improve systems. They should not be free to determine, alone, what the public is entitled to know when those systems affect employment, housing, finance, education, safety or political information.
The case for restraint
The strongest argument against sweeping regulation is not a defense of recklessness. It is a warning about badly designed rules. Artificial intelligence changes quickly, and legislation written around one model or one technical failure may become obsolete before it takes effect. Excessive compliance costs could entrench the largest firms, suppress open-source research and push development toward jurisdictions with weaker safeguards.
Those concerns are real. The Center for Democracy and Technology’s review of 2026 US legislation describes a shift toward targeted measures, including rules for automated decisions, public-sector use, child safety and frontier-model risks. It also notes that federal proposals had not become law during the session covered. The organization’s analysis illustrates both the breadth of the policy problem and the difficulty of producing workable national standards.
Regulators should therefore avoid trying to license every algorithm or freeze a rapidly developing field. A proportionate framework would focus on demonstrable harm, require documentation and testing for high-risk uses, and preserve room for low-risk applications that do not meaningfully affect people’s rights.
Why promises are insufficient
Yet flexibility cannot become an excuse for opacity. A voluntary pledge is only as durable as the commercial incentives behind it. The firms developing frontier systems compete for investment, customers and market share. If safety measures slow deployment while competitors move ahead, a promise without independent verification may become a public-relations asset rather than a meaningful constraint.
International experience points toward a middle path. The European Union has adopted binding rules organized around levels of risk, while the United States has leaned more heavily on executive action and voluntary commitments. China is developing a state-directed framework of its own. Euronews described these as sharply different approaches, each balancing innovation, control and accountability in a different way.
The EU model is not beyond criticism. It can be complicated, expensive and vulnerable to lobbying. But its central principle is sound: obligations should rise with the potential consequences of a system. An AI tool that recommends a playlist should not face the same requirements as one that helps decide whether someone receives a job, a loan or medical care.
What responsible rules would require
Any durable US framework should begin with several modest but essential guarantees:
- Clear disclosure when people are interacting with AI or seeing synthetic media in contexts where deception could matter.
- Independent testing and incident reporting for high-risk systems.
- Human review and a meaningful appeal process when automated decisions affect access to essential opportunities or services.
- Protection for researchers, employees and whistleblowers who document serious safety failures.
- Regular review so rules can adapt without allowing companies to mark their own homework.
These requirements would not settle every dispute. They would, however, establish a floor beneath which competition could not push safety. They would also give smaller companies clearer expectations instead of leaving them to interpret a patchwork of private promises and state laws.
The public debate has been framed too often as a choice between unregulated innovation and bureaucratic paralysis. That is a false choice. The better goal is accountable innovation: fast where the risks are limited, careful where mistakes can permanently damage people’s lives, and transparent enough that citizens can challenge decisions made about them.
Technology companies deserve a voice in writing workable rules. They do not deserve the final word. The promises announced in Washington may be a useful beginning, but only enforceable oversight can turn them into public protection.