Commentary. The argument over artificial-intelligence regulation has reached an odd and revealing stage: some of the companies building the most powerful systems now ask governments to intervene, while others warn that intervention could choke innovation. Our view is that both concerns deserve a hearing—but the public should not be asked to choose between safety and progress when sensible rules can protect both.
The United States is unlikely to pass major AI-safety legislation soon. The political obstacles are substantial, including opposition from President Donald Trump and the difficulty of moving a bill through both chambers of Congress. Yet lawmakers are considering measures ranging from independent reviews to emergency powers that could halt a model posing an imminent catastrophic risk. BBC reporting describes a debate that is active, but increasingly vulnerable to legislative deadlock.
That paralysis is not a neutral outcome. It leaves the public with a patchwork of company promises, state rules and agency interpretations while AI systems spread through schools, workplaces, hospitals, elections and government services. Voluntary commitments may be useful, but they are weakest precisely where commercial incentives are strongest: before a lucrative product launch, after a dangerous failure or when competitors are moving faster.
The industry’s divided message
OpenAI chief executive Sam Altman has argued for a lighter regulatory touch, saying society may have to accept some harms in exchange for the technology’s benefits and people’s freedom to use it. Anthropic chief executive Dario Amodei has taken a more interventionist position, calling for binding safety requirements, third-party testing and government authority to block deployments judged unacceptably risky. The contrast is documented in the Los Angeles Times.
Neither position should be dismissed as self-interest alone. A lighter framework could help smaller firms compete with incumbents, prevent regulators from freezing immature technologies and preserve room for beneficial applications. Heavy licensing requirements might also concentrate power in the largest companies, which can afford compliance departments and influence the rule-making process.
Those are serious objections. Regulation written in panic would be bad policy. A law that treats a spell-checking tool, a medical diagnostic system and an autonomous cyber agent as equivalent would be both costly and absurd. Rules should be proportionate to capability and risk, not to the marketing language surrounding a product.
What the public should demand
But proportionality is not an argument for inaction. It is an argument for precision. Congress should require independent testing for high-risk systems, disclosure of serious incidents, meaningful audit access and clear responsibility when companies deploy models in consequential settings. Regulators should have narrowly defined emergency powers, subject to judicial review and public reporting—not an unlimited veto over software.
Those safeguards would not eliminate mistakes. They would make mistakes visible, assign accountability and create incentives to fix them. The alternative is a system in which the companies that understand the technology best decide how much evidence the public gets to see.
The case for caution is also strengthened by the international character of the problem. AI models cross borders, synthetic media can move faster than election officials and cyber capabilities do not respect national jurisdiction. Britain’s lawmakers have begun examining AI security risks, while scientists and technology executives continue to debate international oversight. The fact that governments disagree does not make common standards unnecessary; it makes them harder—and more urgent—to build.
Our newsroom’s position is straightforward: regulate dangerous capabilities, not innovation itself. Protect research and low-risk uses, but require proof of safety where systems can materially affect liberty, livelihoods, public health or national security. Give smaller companies compliance paths they can afford, and prevent dominant firms from writing rules designed to preserve their advantage.
The strongest counterargument is that regulators may move more slowly than technology. That is true. It is also why rules should establish durable principles—testing, transparency, traceability and accountability—rather than attempt to predict every future product. Cars, aircraft and medicines evolved under regulation; safety standards did not end their development.
AI will continue to advance without waiting for political consensus. The choice is not between regulation and progress. It is between accountable progress and an experiment conducted on everyone else.