Commentary. Artificial intelligence policy has entered the familiar stage of modern regulation: lawmakers are producing rules faster than institutions can enforce them. That is progress of a kind. It is also a warning. Our view is that governments should stop treating the number of AI statutes as evidence of safety and concentrate instead on liability, independent oversight and remedies for people harmed by automated systems.
The regulatory expansion is real. The National Conference of State Legislatures says more than 1,500 AI-related bills were introduced in the United States in 2026, with 280 adopted; disclosure requirements for AI interactions nearly tripled, while deepfake measures became the most common category of enacted laws.[1] In Europe, the European Commission is consulting on enforcement rules for parts of the EU AI Act, including proceedings involving general-purpose AI providers and systems embedded in major platforms and search engines.[2]
Those developments answer a genuine public demand. People deserve to know when they are dealing with a machine, and voters have good reason to worry about synthetic fraud, discriminatory automated decisions and systems that imitate intimacy while exploiting children. The European Union’s risk-based framework places stricter obligations on high-risk uses, including recruitment and worker management, while new rules on algorithmic management are due to be implemented by member states in December 2026.[3]
But disclosure alone is a thin form of protection. A notice that a chatbot is artificial does little for a tenant denied housing, a worker rejected by an opaque screening system or a family targeted by a convincing deepfake. Transparency can become a ritual that shifts responsibility onto users who lack the technical or legal power to challenge the decision.
The stronger approach is straightforward: whoever deploys an AI system in a consequential setting should be able to explain its purpose, document its testing, preserve an audit trail and provide a meaningful human appeal. Developers and deployers should not be able to evade responsibility by pointing at one another. Liability rules should also reflect the reality that many systems change after deployment and that victims rarely have access to the evidence needed to prove what went wrong.
The case against overregulation
The opposing argument deserves more than a dismissive response. Fragmented rules can raise compliance costs, especially for small companies, and may entrench the largest technology firms, which can afford armies of lawyers and auditors. Businesses also warn that vague requirements could delay useful applications in medicine, research and public services. In the United States, federal lawmakers have not yet passed a comprehensive AI law, while states continue to legislate independently, creating a patchwork that companies say is difficult to navigate.[4]
Those concerns are legitimate. Regulation that treats a low-risk productivity tool like an automated benefits decision is bad regulation. So is a compliance regime that rewards paperwork instead of safer products. Rules should be proportionate, technologically neutral and periodically reviewed. Regulators should offer sandboxes and clear standards, not demand permission for every experiment.
Yet the answer to clumsy regulation is better regulation, not regulatory surrender. The EU’s emerging standards-based approach illustrates both the promise and the danger: technical standards can give companies a clearer route to compliance, but they must not become a private substitute for democratic accountability. The Commission has endorsed an AI quality-management standard, while major high-risk obligations have been postponed until at least December 2027.[2] Delay may help businesses prepare; it may also prolong uncertainty for people already exposed to automated decisions.
Our editorial position is therefore neither “ban AI” nor “let innovation decide.” It is that speed should be measured by how quickly institutions can detect harm and repair it, not by how quickly products reach market. Governments should align basic protections across jurisdictions, fund independent regulators and guarantee accessible appeals. Companies that build or deploy powerful systems should publish enough evidence for outside scrutiny without exposing trade secrets or personal data.
AI will not become trustworthy because officials pass another symbolic disclosure law. It will become more trustworthy when the public can identify who made a decision, challenge it, and obtain compensation when the system causes foreseeable harm. The current wave of legislation is an opportunity. It should produce fewer slogans, stronger duties and consequences that reach the people who control these systems.