Commentary. Europe has chosen a politically understandable path on artificial-intelligence regulation: slow the timetable without abandoning the destination. The EU’s Digital Omnibus on AI postpones the application of some high-risk rules, while retaining the bloc’s risk-based framework and adding fixed deadlines for implementation.[1] Our view is that this compromise is defensible—but only if “more time” means serious preparation rather than a quiet weakening of accountability.
The European Parliament and Council reached agreement on the changes in May 2026. Parliament approved the deal in June, the Council adopted it later that month, and the final act was signed and published in July.[1] The agreement sets the application date for rules covering stand-alone high-risk AI systems at 2 December 2027, while high-risk systems embedded in regulated products receive until 2 August 2028.[1]
That delay reflects a real problem. The AI Act’s obligations are technically and administratively demanding. Companies need standards, testing procedures, documentation systems and competent authorities capable of interpreting the law consistently. If rules take effect before those foundations exist, compliance could become a paperwork contest rather than a meaningful check on systems used in employment, education, essential services or safety-sensitive products.
Smaller companies have an especially strong case for caution. Large technology firms can hire compliance teams and legal advisers; smaller developers and deployers may struggle to understand overlapping requirements or to afford repeated assessments while standards remain unsettled. A rushed regime could entrench the biggest incumbents—the opposite of the competition Europe says it wants.
Still, delay has costs that should not be disguised as administrative housekeeping. High-risk AI is not a theoretical category. Automated systems can influence access to jobs, credit, public benefits and other opportunities. Every postponed obligation leaves institutions and affected people with fewer enforceable safeguards. The European Parliament’s own summary says the compromise delays certain requirements but maintains core provisions, including prohibitions on non-consensual intimate or sexual content and child sexual-abuse material.[2] That distinction matters: some harms demand immediate boundaries, regardless of the timetable for more complex conformity assessments.
“The question is not whether Europe should regulate AI, but whether its rules will be usable before they are needed.”
The strongest argument against the delay is therefore not maximalism. It is credibility. Europe presented the AI Act as a model for governing powerful technology through predictable, rights-conscious rules. Repeated extensions would teach businesses and citizens that deadlines are negotiable whenever implementation becomes inconvenient. That would reward lobbying, unsettle investment and weaken public confidence.
The strongest argument for the delay is that bad enforcement can be worse than late enforcement. A regulator without technical capacity may produce inconsistent decisions, while vague standards invite defensive compliance and litigation. The compromise can work if authorities use the extra months to publish concrete guidance, fund national enforcement bodies and provide accessible support for smaller firms. Those steps should be measurable, public and subject to parliamentary scrutiny.
Europe should also resist the temptation to treat innovation and rights as opposing camps. Clear rules can help responsible developers distinguish themselves from reckless ones. But clarity requires more than legislation in the Official Journal. It requires tested standards, transparent audits and remedies that ordinary people can actually use. A system that protects only those able to navigate a complex complaint process is not meaningfully rights-based.
Our position is simple: accept the limited postponement, reject open-ended slippage. The EU should publish a timetable for guidance and accreditation, report progress regularly and preserve the earliest deadlines for clearly defined abuses. If the institutions cannot demonstrate preparation by the new dates, the problem will not be that Europe regulated too much. It will be that Europe promised protection without building the machinery to deliver it.