Commentary. Europe is entering the most consequential phase of its experiment with artificial-intelligence regulation. The European Union’s AI Act is no longer merely a political promise: its transparency obligations began applying on 2 August 2026, while enforcement powers now sit with the EU AI Office and national market-surveillance authorities.[1] Our view is straightforward: the EU should resist the temptation to weaken the law further before regulators have had a fair chance to enforce it.
That position does not require pretending the original timetable was flawless. The rules are technically demanding, especially for smaller companies that must identify affected systems, document data and risk controls, and explain when users are interacting with synthetic content. The EU’s July AI Omnibus already simplified parts of the framework, expanded testing opportunities and postponed several high-risk obligations.[1] Those changes may reduce needless compliance costs. They should not become a permanent escape route.
The case for patience
Businesses have a legitimate complaint: regulation written faster than the technology evolves can produce uncertainty rather than safety. The most demanding rules for high-risk systems covered by Annex III are now scheduled to apply from December 2027, while certain systems governed by sectoral safety legislation receive an August 2028 deadline.[1] Companies argue that they need time to build conformity assessments, find qualified auditors and understand guidance that is still developing.
That argument matters. A compliance system that only large firms can afford could entrench the biggest technology companies, precisely the opposite of the competition Europe says it wants. Delaying complex obligations can also prevent regulators from enforcing vague rules inconsistently. In that sense, the Omnibus is not necessarily deregulation; it may be an attempt to make regulation workable.
There is a further concern about strategic disadvantage. While Europe builds a detailed rulebook, the United States is pursuing a more permissive and fragmented approach, and policymakers elsewhere are emphasizing domestic AI development. Excessive friction could push investment and research away from European markets.
Why delay must have limits
Yet the counterargument is stronger where AI affects people’s rights, livelihoods and access to essential services. The EU itself classifies systems used in areas such as employment, education, critical infrastructure, law enforcement and essential services as high-risk categories. The fact that compliance is difficult does not make the underlying risks less immediate.
Transparency rules are therefore an important first test. The framework requires providers and deployers to address disclosure obligations for AI systems, including synthetic or manipulated content.[2] Users should not have to guess whether an image, voice or text was generated or altered by a machine, particularly during elections, emergencies or financial transactions. Transparency cannot solve every problem, but secrecy makes accountability nearly impossible.
“Core transparency obligations under Article 50 will still apply from 2 August 2026.”
That conclusion is supported by legal analysis from Stephenson Harwood, which notes that the Article 50 requirements remain in force despite postponed deadlines elsewhere.[3]
The danger now is regulatory drift: each delay is presented as technical housekeeping, but together the postponements could teach companies that political pressure will soften any rule that becomes inconvenient. Regulators should publish clear guidance, offer practical support to smaller firms and prioritize the most serious harms. They should not quietly convert a rights-based system into a voluntary code.
Enforcement is the credibility test
Europe’s policymakers should adopt three disciplines. First, distinguish genuinely low-risk experimentation from systems that make or influence consequential decisions about people. Second, measure compliance by outcomes, not by the volume of paperwork. Third, report publicly on enforcement, including investigations, remedies and the reasoning behind them.
Industry deserves predictable rules. The public deserves protection from undisclosed manipulation, discriminatory automated decisions and systems deployed before their risks are understood. These demands are not mutually exclusive. Clear standards and proportionate enforcement can give responsible companies an advantage over those that cut corners.
Our editorial position is not that Europe should regulate every new model into paralysis. It is that the EU should stop treating enforcement as an obstacle to innovation. Trust is infrastructure for adoption. If citizens believe AI is hidden, unaccountable or imposed on them, backlash will be more damaging to the technology’s future than sensible safeguards would be.
The AI Act’s next chapter should therefore be judged by implementation, not ambition. Europe has already written the rules. Now it must show that they apply to powerful companies, smaller developers and public institutions alike.