Commentary. Europe’s revised artificial-intelligence rulebook is a reasonable attempt to reconcile two legitimate concerns: the need to protect people from powerful systems and the risk that cumbersome regulation will leave European companies behind. But the compromise now entering force should be judged by what it enables regulators to enforce—not simply by how much paperwork it removes.

The European Union’s AI Omnibus, which entered into force on 27 July 2026, simplified compliance, expanded testing opportunities and extended deadlines for some high-risk systems. The revised timetable postpones rules for certain stand-alone high-risk applications until 2 December 2027, while high-risk systems embedded in products receive until 2 August 2028.[1]

Our view is straightforward: the delay is defensible, but only if it is used to build enforcement capacity rather than to postpone political responsibility. A regulation that arrives late but works is preferable to one that arrives on time and collapses under ambiguity. A delay that merely gives companies more time to lobby, experiment and scale without effective scrutiny would be a different matter.

The case for simplification

Businesses have a strong argument. The original AI Act created overlapping obligations for companies that may lack the legal teams and technical staff of the largest technology firms. The new package offers simplified obligations for smaller and mid-cap companies, expands regulatory sandboxes and seeks to clarify how the AI Act interacts with sector-specific product rules.[1]

Those changes could make compliance more predictable. They may also help European developers test useful systems in health, education, manufacturing and public services without navigating a regulatory maze designed primarily around worst-case scenarios. The European Parliament and Council also added prohibitions covering non-consensual intimate or sexual content and child sexual abuse material, showing that simplification need not mean abandoning hard boundaries.[2]

Europe’s competitiveness concern is not imaginary. If firms face high compliance costs before they have the scale to absorb them, investment and talent can move elsewhere. Regulation that protects citizens but leaves them dependent on foreign providers may ultimately weaken democratic control over technology.

The case against complacency

Yet postponing high-risk obligations also postpones protections in precisely the areas where errors can have lasting consequences. The delayed categories include systems used in sensitive settings such as employment, education, migration, critical infrastructure and biometric identification.[1] In these environments, a flawed automated recommendation can affect a person’s livelihood, liberty or access to essential services.

The European Commission’s scientific and ethics advisers have warned that AI oversight must preserve human agency, maintain meaningful human supervision and develop independent evaluation capacity rather than relying solely on commercial developers.[3] That warning should be treated as the test of the revised regime. Formal compliance documents are not enough if neither public authorities nor affected individuals can determine whether a system is reliable, discriminatory or manipulable.

Critics of strict regulation will say that these risks are already addressed by existing laws governing privacy, discrimination, consumer protection and product safety. They are partly right. Europe should not create redundant rules that produce contradictory duties. But existing law often acts after harm occurs, while AI governance must also examine systems before deployment and monitor them as conditions change. A model can be legally purchased, technically impressive and still unsuitable for a consequential decision.

What Europe should do now

The answer is not to reverse the Omnibus. It is to attach measurable obligations to the breathing room it creates. Governments should publish clear implementation guidance, fund independent testing and require timely reporting of serious failures. Regulators should be able to inspect systems in practice, not merely review assurances supplied by vendors. Companies, meanwhile, should disclose enough about performance and limitations for users to make informed decisions without exposing legitimate trade secrets.

Europe should also resist the temptation to equate innovation with deregulation. The best competitive advantage may be trustworthy deployment: systems that hospitals, employers and public agencies can adopt because their risks are understood and their failures can be challenged. That requires rules that are proportionate, comprehensible and consistently enforced.

The AI Act’s revised calendar offers Europe a second chance to get the balance right. The continent should use it neither to punish experimentation nor to excuse inaction. Our editorial position is that safety must remain the condition of scale. If Europe can simplify the path to compliance while strengthening independent oversight, the compromise will look prudent. If it merely delays scrutiny, it will look like surrender dressed as pragmatism.

Sources