Commentary. The world is entering an awkward phase of artificial-intelligence governance: governments are moving faster than before, yet not necessarily in the same direction. That should concern anyone who believes powerful technology needs rules that are understandable, enforceable and capable of surviving political fashion.
In the United States, the federal government is preparing new requirements for agencies buying AI systems, while lawmakers from both parties are proposing sharply different approaches to liability and oversight. At the same time, several states are advancing their own rules, creating a patchwork that companies may find costly and citizens may find confusing. [Federal News Network] [Washington Examiner]
Europe is taking a more institutional route. The European Commission has opened consultations on children’s online safety and the use of copyrighted material in AI, while a first harmonised standard supporting implementation of the EU AI Act has been published. Singapore, meanwhile, has issued AI risk-management guidelines for financial institutions, with implementation scheduled for next year. [Simmons & Simmons] [EU AI Act Newsletter]
The case for moving carefully
Our view is straightforward: regulation should begin with demonstrable risks, not with grand theories about the future. Governments should require clear documentation of training data and system limits, independent testing for high-risk uses, meaningful appeal rights when automated decisions harm people, and disclosure when the public is interacting with synthetic media.
Those measures would not require governments to predict whether machines will become “superintelligent.” They would address problems already visible: unreliable automated decisions, opaque procurement, impersonation, privacy risks and failures in sectors such as finance, health and education. They would also give responsible companies a common baseline instead of rewarding those most willing to exploit uncertainty.
“The public interest demands rules that can be tested in court, audited by experts and understood by ordinary users.”
That does not mean every proposed restriction is wise. Broad bans on systems defined by vague capabilities could freeze legitimate research and place smaller firms at a disadvantage. Heavy compliance costs may entrench the largest technology companies, which can afford lawyers, auditors and dedicated policy teams. And lawmakers who impose detailed technical requirements too early may lock regulation to a fast-changing technology.
Innovation is not a rebuttal
Industry’s strongest argument is that excessive regulation could slow productivity and allow less cautious countries to gain an economic or military advantage. That risk is real. The emerging policy debate includes proposals ranging from criminal and civil liability for particular harms to more sweeping restrictions on systems that exceed human performance. Such proposals deserve scrutiny rather than reflexive approval. [Washington Examiner]
But “innovation” cannot become a blanket exemption from accountability. A company that deploys an AI system into a hospital, bank or public agency is not merely releasing a clever piece of software; it is influencing decisions about people’s lives. The relevant question is not whether regulation slows some deployment. It is whether the deployment is safe enough, transparent enough and reversible enough to justify the risk.
The most credible framework would therefore be proportionate. Low-risk creative tools should face light-touch duties. Systems used in employment, credit, policing, medicine or essential public services should face stronger testing, record-keeping and human oversight. Regulators should be able to impose temporary restrictions when serious failures occur, while preserving routes for appeal and independent review.
A chance for democratic discipline
The United States, Europe and other jurisdictions do not need identical laws. They do need compatible principles. Divergent rules can become a laboratory for better policy, but only if governments share evidence about failures, coordinate on testing and resist turning AI standards into trade weapons.
For now, the danger is less that one government will regulate too much than that many governments will regulate inconsistently, while companies market speed as proof of safety. Public trust will not be restored by new terminology or competing slogans. It will be earned through rules that identify responsibility before harm occurs and enforce consequences when safeguards fail.
That is the standard this newsroom supports: not a moratorium on progress, and not a race to the lightest rules, but accountable innovation. The technology may be new. The civic obligations are not.