Commentary. The argument over artificial-intelligence regulation has acquired a familiar shape: technology companies warn against rules that could slow innovation, while researchers and civil-society groups warn that waiting for certainty could leave the public absorbing the costs of failure. Our view is that governments should regulate advanced AI now—not to freeze an emerging industry, but to establish minimum obligations before the technology becomes too consequential to govern.
The case for action is no longer hypothetical. Experts cited in recent coverage have pointed to incidents involving AI agents behaving unpredictably, while prominent researchers have called for safeguards that go beyond voluntary promises by technology companies. Around 400 scientists, including Geoffrey Hinton and Yoshua Bengio, recently backed an international body focused on containment standards and reporting incidents involving advanced cyber-capable systems. CBC News reports that supporters of oversight also include Canadian Prime Minister Mark Carney and other public figures.
That does not mean every AI product belongs under the same regulatory regime. A spell-checking tool, a medical diagnostic system and an autonomous agent capable of accessing computer networks present different levels of risk. Sensible policy must distinguish between them. Rules should become more demanding when systems can affect health, finances, elections, critical infrastructure or personal safety.
The strongest argument against haste
Opponents of broad regulation have legitimate concerns. AI development is moving quickly, and a law written around today’s systems could become obsolete before it takes effect. Compliance costs could also entrench the largest companies, forcing smaller laboratories and start-ups to spend money on lawyers and audits rather than research. Heavy-handed restrictions imposed by one country could push development elsewhere without reducing global risks.
There is also a real danger in treating expert consensus as a substitute for democratic judgment. Scientists can identify technical hazards, but elected institutions must decide which risks society is willing to accept, who should pay for mitigation and how rights will be protected. Regulation designed in closed rooms would deserve public suspicion, even when its intentions are admirable.
These objections argue for flexible rules, not regulatory paralysis. Governments can require testing, documentation, incident reporting and independent audits for high-risk systems while allowing lower-risk applications to develop under lighter obligations. Standards can be reviewed regularly, and regulators can be given technical expertise rather than relying solely on industry assurances.
Self-policing is not enough
The central weakness of voluntary governance is structural. Companies are being asked to restrain products that determine their valuations, market share and competitive position. Many firms may act responsibly, but a system that depends on universal restraint will fail when one important actor decides that speed matters more than caution.
Nor should regulation be reduced to a contest between innovation and safety. Clear rules can help serious companies compete by establishing a baseline that applies across the market. They can also give the public a way to seek redress when automated decisions cause harm. The alternative—leaving every dispute to private contracts and after-the-fact lawsuits—places too much power in the hands of firms whose systems ordinary people cannot inspect.
Build guardrails that can travel
AI is global, so national rules alone will be inadequate. International coordination will be difficult, especially where governments view advanced AI as an economic or military advantage. Still, common requirements for incident disclosure, model evaluation and safeguards against cyber misuse would be a practical starting point. The goal need not be a single world regulator; it should be compatible expectations that prevent the weakest jurisdiction from becoming the default refuge for dangerous practices.
The right response is neither panic nor faith in corporate benevolence. It is accountable, proportionate oversight: strong enough to address systems with the capacity to cause serious harm, adaptable enough to keep pace with research, and transparent enough for the public to challenge it. Innovation deserves room to grow. It does not deserve immunity from rules.