Commentary. The latest White House effort to bring artificial-intelligence companies into a common safety framework is welcome. It is also insufficient. The central question is no longer whether AI developers should acknowledge risk; it is whether the public should have to trust companies to police themselves while their systems become more capable and more deeply embedded in daily life.
Tech executives meeting with President Donald Trump reportedly endorsed a “morally binding” commitment to build safeguards. The resulting White House Accord leaves open the possibility that some commitments could eventually become law or regulation, according to SBS News. That language matters. It recognizes that voluntary standards may be a bridge to formal oversight rather than a permanent substitute for it.
Our view is straightforward: the United States should regulate high-risk AI, but it should do so narrowly, transparently and in a way that preserves room for useful innovation. A blanket moratorium would be blunt and potentially self-defeating. A voluntary pledge with no meaningful inspection or penalty would be worse: reassuring in public, weak in practice.
The case for restraint
The strongest argument against heavy regulation is not a caricature of corporate self-interest. AI development is international, fast-moving and economically consequential. Policymakers who impose costly or slow approval systems could encourage investment and talent to move elsewhere, while competitors—particularly China—continue developing advanced systems. That concern has been voiced by administration officials and technology leaders who favor safety measures without a broad pause, as 41NBC reported.
Rules written for one generation of models may also become obsolete quickly. Prescriptive technical mandates can burden smaller firms more than dominant companies, entrenching the very giants regulators claim to oversee. Lawmakers should therefore resist regulating every chatbot feature or treating speculative predictions as established fact.
Why promises are not enough
But speed is not a policy principle. Nor is national competition a reason to make citizens the testing ground for systems whose failures can affect elections, workplaces, critical infrastructure and personal safety.
Researchers and policymakers have called for independent testing, greater transparency and common safety standards. A separate report in The Straits Times described proposals including embedded independent evaluators and international coordination. These are not demands to halt all progress. They are basic mechanisms for determining whether developers’ assurances deserve public trust.
The political deadlock is real. The BBC reported that comprehensive US AI legislation faces long odds, while Reuters described Senate discussions over a possible “duty of care” requiring companies to show they took reasonable precautions against catastrophic harm. Even a modest bill could fail if the White House opposes new federal controls.
“Light touch” should mean proportionate oversight—not the absence of oversight.
A workable middle ground
Congress should begin with a limited framework focused on frontier models and clearly defined high-risk uses. Developers should disclose evaluation methods, report serious incidents, preserve audit records and demonstrate safeguards before releasing systems capable of significant physical, financial or cyber harm. An independent public authority should have access to testing data and the power to investigate failures.
That approach would answer the industry’s legitimate concerns better than sweeping restrictions. It would establish predictable rules, reward companies that invest in safety and prevent responsible firms from being undercut by competitors willing to conceal problems. Small developers could receive tailored obligations, while the largest and most capable systems would face the strongest scrutiny.
Voluntary agreements still have a role. They can move faster than legislation and help establish technical norms. But they should be treated as pilots with deadlines, public reporting and a clear path to enforceable standards. No industry should be asked to grade its own homework indefinitely when the consequences of failure fall on everyone else.
AI may deliver enormous benefits. That is precisely why public confidence matters. The choice is not innovation or safety; it is whether safety becomes part of innovation’s foundation or an emergency repair attempted after preventable damage. The White House accord is a useful opening. The next step must be rules that are independent, targeted and enforceable.