Commentary. The argument over artificial-intelligence regulation is often presented as a choice between innovation and safety. That framing is too convenient. The real choice before lawmakers is whether the public will receive clear, enforceable protections as AI moves from novelty into decisions about work, money, health and identity.
Recent developments show why the issue can no longer be postponed. The European Union’s Digital Omnibus on AI entered into force in July, while the European Commission opened a consultation in October on procedures for investigating and penalising certain AI providers under the bloc’s AI Act. Singapore has also issued final AI risk-management guidelines for financial institutions, with core governance requirements due to begin in October 2027. In California, lawmakers amended the state’s AI transparency regime to broaden its coverage and require provenance disclosures for content altered by publicly accessible generative-AI systems.
These measures are not identical, and that is precisely the problem. Businesses operating across borders face a growing patchwork of definitions, deadlines and documentation requirements. Smaller companies may struggle to understand which rules apply, while larger firms can absorb compliance costs that competitors cannot. The risk is not merely bureaucratic. Excessively complicated regulation can push useful tools out of smaller markets or entrench the biggest technology companies, which are best equipped to hire lawyers and auditors.
Those objections deserve a serious hearing. AI systems can improve medical research, automate routine work and help people communicate across languages. A regulator that treats every model as equally dangerous would waste scarce public resources and could slow beneficial experimentation. Financial supervisors, for example, need rules proportionate to the risks of a trading algorithm or a customer-service assistant, rather than a single universal checklist.
But proportionality should not become an excuse for opacity. When an automated system contributes to a consequential decision, the person affected should know that AI was involved, understand the practical reason for the outcome and have a meaningful route to challenge it. Those are not anti-technology demands. They are ordinary standards of due process, translated for an automated age.
Transparency must be usable
Our view is that regulators should focus first on a small number of rights that can be explained in plain language and enforced consistently. Disclosure should tell people when they are interacting with an AI system or when synthetic media has been materially altered. High-impact decisions should trigger documented testing, human accountability and an appeal process. Independent oversight should have access to enough information to investigate—not necessarily every trade secret, but more than polished assurances from vendors.
The emerging rules point in this direction, but they also risk making transparency a technical exercise. A label hidden in metadata will not help a job applicant who receives no understandable explanation for rejection. A risk assessment filed with an authority will not restore trust if the affected public cannot learn who is responsible. Regulation should measure whether people can exercise their rights, not merely whether a company has completed a form.
There is also a legitimate concern that disclosure tools can be unreliable. Synthetic-content markers may be stripped, detection systems may produce false positives, and provenance standards may not work uniformly across platforms. Policymakers should therefore avoid promising certainty they cannot deliver. Labels should be treated as one layer of accountability, alongside records, audits and remedies.
The best path is neither laissez-faire nor a prohibitionist reflex. Governments should coordinate core definitions, align reporting requirements and give small organizations practical compliance support. They should also preserve room for research and low-risk uses while applying stricter duties to systems that influence access to employment, credit, housing, education or essential services.
AI will not become trustworthy because officials publish more rules, nor because companies repeat that their systems are safe. Trust will come when an ordinary person can identify automation, obtain a comprehensible explanation and challenge a consequential error. That is the public promise regulation should make—and the standard by which new laws should be judged.