Commentary. The argument over artificial-intelligence regulation has moved beyond whether rules are necessary. The more urgent question is whether governments can build safeguards quickly enough to matter without suffocating a technology that may deliver substantial economic and social benefits.
Our view is clear: voluntary commitments should remain part of the framework, but they cannot be the framework. Systems that can influence employment, credit, education, health care and public information should not be governed principally by promises from the companies that develop and sell them.
That position is not a rejection of innovation. It is a recognition of incentives. Technology companies compete to release more capable products, attract investment and win customers. Even conscientious executives face pressure to move quickly. Asking those same firms to decide how much risk society should accept is an inadequate substitute for democratically accountable rules.
The case for stronger oversight
The European debate illustrates the stakes. Members of the European Parliament have recently rejected reliance on self-regulation and called for international action, including discussion of a binding global AI treaty. They have also revived concerns about liability, arguing that people harmed by AI systems may lack clear routes to compensation under existing law. The Center for Democracy and Technology Europe reports that these debates are unfolding alongside negotiations on changes to the EU’s digital rules.
This is a sensible direction. Regulation should require developers to test high-risk systems, document their data and design choices, report serious incidents and preserve meaningful human oversight. It should also make responsibility legible. When an automated system causes harm, victims should not have to navigate a maze of contractors, cloud providers and opaque terms of service before discovering who is accountable.
Recent policy discussions in Britain point to the same weakness. A legal analysis describes the country’s existing sector-by-sector approach as increasingly contested, while reporting on a parliamentary human-rights review characterizes the framework as “patchy and confused.” The analysis says the review called for rules covering AI from development through deployment and use.
The counterargument deserves respect
Critics of broad regulation are not simply defenders of corporate impunity. They warn that poorly designed rules could entrench the largest companies, raise compliance costs for smaller firms and slow useful applications in medicine, science and public administration. They also argue that existing consumer-protection, competition and sector-specific laws can address many harms without creating a sprawling new bureaucracy.
Those concerns are real. A rule written around today’s models could become obsolete before it takes effect. Legislators may also mistake spectacular hypothetical risks for the problems people face now, such as discriminatory automated decisions, fabricated information, insecure software and the misuse of personal data.
That is why the answer should not be a single prescriptive code covering every algorithm. Regulation should be proportionate to capability and consequence. A text-generation tool used for brainstorming should not face the same obligations as a system screening job applicants, approving loans or directing critical infrastructure.
What responsible rules would look like
Governments should focus on outcomes rather than trying to dictate one technical method. Independent testing, audit access, incident reporting and clear appeal mechanisms are more durable than detailed requirements tied to a particular model architecture. Regulators also need technical expertise, adequate funding and the authority to impose meaningful penalties.
International coordination matters, too. AI systems cross borders, and divergent rules can encourage companies to move risky development to jurisdictions with weak oversight. Yet global agreement should not become an excuse for delay. Countries can establish domestic protections while working toward compatible standards.
The industry’s own proposals should be welcomed, but treated as evidence for regulation rather than a replacement for it. If companies can agree on safety benchmarks and independent evaluation, those practices can inform binding requirements. If they cannot, that failure strengthens the case for public intervention.
The choice is not innovation or regulation. It is whether innovation proceeds inside a system that earns public trust. The public should not be asked to accept irreversible risks in exchange for promises that were never enforceable. Sensible, adaptable rules would give useful AI room to grow—and give citizens a meaningful say in how that growth affects their lives.