Commentary. The argument over artificial-intelligence regulation has reached a familiar and dangerous stage: companies warn that rules will slow innovation, governments hesitate, and the public is asked to trust assurances that are difficult to test. Our view is straightforward: AI developers should not be allowed to set the terms of accountability themselves.

The issue is current and consequential. The European Union has shelved proposed AI liability rules, a move that has revived debate over whether people harmed by automated systems will have practical ways to seek redress. At the same time, European officials say work toward an international AI-safety framework will continue despite opposition from Washington.[1][2]

That tension deserves more than slogans. AI systems are already being used in settings where mistakes can affect employment, health, credit, education and public services. When responsibility is divided among a developer, a deployer and an end user, victims can face an exhausting legal puzzle: which system failed, who knew about the risk, and who had the power to prevent the harm?

The case for caution

Industry’s counterargument is not frivolous. AI products vary enormously, from general-purpose models to highly specialized medical tools. A rigid liability regime could encourage defensive design, raise costs for smaller companies and make developers reluctant to release useful systems. Policymakers also risk writing rules around today’s technology while the underlying systems change faster than legislation.

Those concerns are especially serious for Europe, where companies already navigate a dense regulatory environment. The European Commission’s decision to delay or abandon liability proposals reflects a broader fear that compliance burdens could weaken the region’s competitiveness. In the United Kingdom, financial regulators have likewise cautioned that regulation may not be the right place to begin.[2][3]

But “move carefully” must not become “do nothing.” Existing legal remedies are often too slow, expensive or technically demanding for individuals. Voluntary commitments can improve safety, but they remain vulnerable to commercial pressure and leadership changes. A promise is not the same thing as a duty enforceable by an independent authority.

What proportionate accountability looks like

Good regulation need not treat every algorithm as a hazardous machine. It should focus on risk and control. Developers of powerful general-purpose systems should document training and testing, preserve meaningful audit records and disclose known limitations. Organizations deploying AI in high-impact decisions should be required to provide notice, human review and a usable appeals process.

Liability should also follow practical control. A company that designs and markets a system should not escape responsibility merely because a customer deploys it in a predictable way. Conversely, a customer that ignores warnings or uses a tool outside its stated purpose should not be able to shift every consequence onto the developer.

Independent testing is essential. Recent reporting on government and corporate AI oversight shows why public officials are demanding clearer answers about catastrophic risks, safety and accountability from major firms.[4] The answer cannot be to place all sensitive information in the hands of competitors or bureaucrats. Confidential audits, regulator access and protections for whistleblowers can balance security with transparency.

Speed is not the only public interest

The strongest objection to regulation is that it could slow beneficial innovation. That may happen at the margins. Yet speed has never been the sole standard for technologies that can shape people’s lives. Aviation, pharmaceuticals and financial markets all developed rules because the consequences of failure were distributed across the public, not confined to the inventor.

The same principle should govern AI. Europe should refine its approach rather than abandon accountability, while the United States should resist treating every safeguard as an attack on innovation. International coordination would reduce fragmentation, but it should not become an excuse to wait indefinitely for consensus.

Our position is not anti-technology. It is pro-choice, pro-evidence and pro-redress. The public can accept uncertainty when institutions admit it, measure it and provide remedies when systems fail. What it should not accept is a future in which companies claim the benefits of automation while everyone else absorbs the risks.

Sources