Commentary. Europe’s artificial-intelligence rulebook has entered its more consequential phase. The EU AI Act’s enforcement provisions for prohibited practices began applying on 2 August 2026, according to the European Commission, while national authorities are still being designated and operational details continue to develop. The Commission’s implementation report makes clear that adopting a law is not the same as making it work.
Our view is straightforward: Europe should enforce the Act firmly, but intelligently. The answer to uncertainty should not be to abandon safeguards. Nor should it be to impose such rigid or confusing obligations that smaller companies, public institutions and researchers are pushed out of the market.
A necessary brake on unchecked deployment
The case for enforcement is stronger than ever. AI systems increasingly influence employment, education, policing, credit and access to services. Where these systems are opaque or unreliable, individuals may have little practical ability to challenge an outcome. Rules on prohibited uses, transparency and high-risk systems create a baseline that voluntary promises cannot reliably provide.
That baseline also serves businesses. Clear requirements can reward companies that invest in testing, documentation and security rather than those that cut corners. A common European framework is preferable to a patchwork of national regimes, provided that regulators apply it consistently.
Yet the Act’s supporters should acknowledge the costs. Compliance can be expensive, particularly for smaller firms that lack specialist legal and technical teams. Ambiguous classifications may encourage defensive over-compliance, delaying useful products while leaving genuinely dangerous applications untouched. The risk is not imaginary: the EU and its member states have already been adjusting implementation timelines and simplifying some requirements through a 2026 “Digital Omnibus” regulation. The amended regulation reflects pressure to make the framework more workable.
Simplification must not become dilution
There is a legitimate argument for reducing duplication. Companies should not have to submit near-identical information to multiple authorities, and regulators should not demand paperwork that contributes little to safety. Shared technical standards, predictable guidance and a single clear channel for incident reporting would improve compliance without weakening protections.
But simplification becomes dangerous when it means postponing accountability. The European Commission says the Act’s enforcement machinery is still being assembled, even though key prohibitions are already applicable. That gap should be closed through staffing, expertise and coordination—not by allowing companies to treat the rules as optional.
Regulators should publish practical examples, prioritize high-impact deployments and distinguish accidental technical mistakes from deliberate concealment. Fines alone will not build trust. Effective enforcement also requires audits, accessible complaints procedures and remedies that ordinary people can use.
The political temptation to choose speed
Europe faces intense pressure to compete with the United States and China in AI. Industry groups warn that regulation could slow investment, encourage companies to relocate and deprive European consumers of new tools. Those concerns deserve consideration, especially when other jurisdictions are moving quickly and regulatory requirements differ across borders.
Still, speed is not a strategy. A market that rewards unreliable systems may produce rapid adoption followed by costly failures, lawsuits and public backlash. Trust is an economic asset, and it cannot be manufactured after a preventable harm has occurred.
The better choice is a regulatory model that is strict about outcomes and flexible about methods. Companies should be able to meet safety goals through different technical approaches, while authorities should retain the power to intervene when systems produce discriminatory, deceptive or dangerous effects.
Europe’s AI experiment will not be judged by the elegance of its legislation. It will be judged by whether people can understand when AI is being used, challenge consequential decisions and obtain redress when systems fail. Enforcement is the point at which principles become real. The EU should simplify the bureaucracy, not surrender the protections.