Commentary. The argument over artificial-intelligence regulation has moved beyond whether governments should act. The more consequential question is whether they can build rules that work in the places where AI already carries real power: finance, employment, education, health care and public services.
Recent developments suggest that the regulatory push is gathering pace. The European Commission has opened a consultation on procedures for investigating and penalising providers of certain systems under the EU AI Act, while Singapore’s financial regulator has issued guidelines for institutions using AI. The Financial Stability Board has also called for closer monitoring of autonomous AI in financial institutions. These are not abstract debates. They concern systems that may influence credit decisions, move money or act with limited human supervision.
Our view is straightforward: governments should regulate high-impact uses of AI now, but they should resist the temptation to regulate every model, feature or speculative risk in the same way. A rulebook that treats a customer-service assistant like an autonomous trading agent would be both inefficient and politically brittle.
The case for firm oversight
The strongest argument for regulation is accountability. When an automated system denies a loan, misidentifies a person or gives dangerous advice, affected individuals need more than an apology from a company. They need a clear explanation of who was responsible, what evidence was used and how a decision can be challenged.
Financial services offer an especially convincing case. AI can increase efficiency, but failures can spread quickly through interconnected institutions. The Financial Stability Board’s call for closer monitoring of autonomous systems reflects a sensible principle: the more independently a system can act, the stronger the controls around testing, logging, human intervention and incident reporting should be.
Public support also appears to be broadening. A Reuters/Ipsos poll cited by CIO Dive found that majorities of both Democratic and Republican voters supported stricter federal AI regulation, while another survey reported that 66% of registered voters wanted more oversight. Those figures do not settle policy questions, but they do undermine the claim that regulation is merely a partisan or anti-technology project.
The counterargument deserves respect
Opponents are right to warn that poorly designed rules could entrench the largest technology companies. Compliance costs, complex certification procedures and uncertain liability may be manageable for a global platform but prohibitive for a small research group or start-up. Regulation written too broadly could also push development into less transparent jurisdictions.
There is a second legitimate concern: technology changes faster than legislation. A law that names particular techniques or product categories may become obsolete before agencies finish implementing it. Policymakers should therefore prefer performance-based obligations—such as accuracy, security, documentation and avenues for appeal—over rigid technical mandates wherever possible.
Innovation is not a trivial public good. AI may improve medical research, accessibility tools and productivity. A regulatory system that makes experimentation impossible would sacrifice benefits that are difficult to measure in advance.
What responsible rules should require
That is why the best approach is neither a free-for-all nor blanket prohibition. It is a tiered system built around consequences.
- Require independent testing and meaningful human oversight for high-impact systems.
- Give people notice when automated tools materially shape decisions about them.
- Make companies preserve records sufficient to investigate failures.
- Impose rapid reporting duties for serious incidents and security breaches.
- Provide small firms with clear, proportionate compliance pathways.
- Give regulators technical expertise and resources to enforce the rules.
The last point is the one most often neglected. A right without a remedy is a slogan. Europe’s AI Act, Singapore’s financial guidance and proposed measures elsewhere will matter only if regulators can inspect systems, impose penalties and compel corrective action. Consultations and frameworks are useful; enforcement is what changes behaviour.
AI policy should also be judged by outcomes rather than headlines. Governments should publish anonymised enforcement data, audit results and evidence about whether rules reduce harm. Where a requirement fails, it should be revised openly. Where companies demonstrate safer methods, regulators should reward that progress rather than preserve obsolete paperwork.
The choice is not between innovation and regulation. It is between accountable innovation and innovation whose costs are quietly imposed on people with the least power to resist them. The public can accept technological risk when responsibility is visible and redress is real. It should not be asked to accept systems that make consequential decisions while everyone involved disclaims responsibility.