Commentary: At Archange Shadows, we believe the latest turn in artificial-intelligence policy gets one thing right and another dangerously wrong. Governments should make rules workable. They should not confuse simplification with safety.
The European Union’s revised AI framework, adopted this year as part of the so-called Digital Omnibus, delays several obligations for high-risk systems and postpones requirements to identify AI-generated content.[1] The Council of the EU says the changes are intended to streamline and simplify the rules.[2] That is a reasonable objective. Regulation that is too complex, too expensive or impossible to interpret can protect established firms while shutting out smaller competitors.
Yet postponement is not neutral. It changes who carries the risk while regulators continue to build enforcement capacity. In the meantime, people may still encounter automated systems in hiring, education, health care, policing and credit—areas where a faulty or discriminatory decision can affect a life long before a formal compliance deadline arrives.
The case for a lighter touch
Supporters of deregulation argue that Europe is moving too slowly while the United States and China compete to develop and deploy increasingly capable systems. Companies need predictable rules, they say, not overlapping reporting duties that divert money and talent from research. Washington’s more hands-off approach reflects precisely that concern: fewer federal barriers, in this view, mean faster innovation and stronger competitiveness.
Businesses also face a genuine patchwork. States in the United States are adopting their own AI laws, while international requirements continue to multiply. One recent industry analysis described the result as more rules in practice even where national governments promise deregulation.[3] That fragmentation can make responsible compliance harder, not easier.
These arguments deserve more than caricature. A regulation that cannot be understood or enforced is a poor regulation. Policymakers should distinguish between a small company using an established model for routine administrative work and a developer releasing a frontier system with broad, difficult-to-predict capabilities.
Why delay still has a cost
But the answer cannot be to postpone safeguards without a credible interim standard. The EU’s revised timetable places deadlines for some high-risk obligations in 2027 and 2028, while content-marking requirements are also delayed for certain systems.[1] A government can call that breathing room; citizens may experience it as a period in which accountability is optional.
California’s proposed approach illustrates the opposite instinct. Governor Gavin Newsom has called for expanded oversight of advanced models, including independent safety monitors and emergency powers to halt systems judged dangerous.[4] Critics will understandably question whether officials can assess rapidly evolving technology well enough to intervene, or whether emergency authority could become a political weapon. Those concerns are real. So is the concern that voluntary promises will fail when commercial pressure rises.
Our view is that governments should regulate consequences, not hype. They should require clear documentation, meaningful testing, incident reporting and human accountability when AI is used in high-impact decisions. They should also offer safe harbors for firms that meet transparent standards, reducing the incentive to avoid innovation through legal uncertainty.
A workable compromise
The best policy would combine a firm floor with flexible implementation. Basic rights—notice, explanation, privacy, appeal and protection against discriminatory outcomes—should not depend on whether a deadline has been extended. Technical rules can evolve, but these principles should remain stable.
Regulators should publish plain-language guidance, coordinate across borders and give smaller developers access to testing resources. Enforcement should focus first on demonstrable harm and reckless conduct, not paperwork errors. At the same time, companies should not be allowed to use complexity as an excuse for withholding information from the people affected by their systems.
Europe’s simplification effort is therefore neither a triumph nor a collapse. It is a test. If delay is used to clarify rules and build competent oversight, it may improve the law. If it becomes a quiet retreat from responsibility, the public will pay the price—and businesses will inherit a less predictable market when trust breaks down.
Speed matters. So does knowing where the brakes are.