Commentary. Europe is preparing to delay some of the most demanding obligations in its Artificial Intelligence Act. The argument is familiar: companies need more time, regulators need clearer technical standards, and Europe cannot afford to burden its emerging AI sector while the United States and China move faster. Those concerns are legitimate. But the proposed timetable risks turning a temporary implementation problem into a permanent political habit—promising safety later, after the technology has already spread.
The European Parliament’s legislative monitoring page says the Commission’s Digital Omnibus proposal would postpone obligations for high-risk, stand-alone AI systems until December 2, 2027, and for high-risk AI embedded in regulated products until August 2, 2028.[1] The proposed changes do not erase every deadline: certain prohibitions and transparency requirements would still take effect earlier.[2]
That distinction matters. A delay is not automatically deregulation. High-risk systems—those used in settings such as employment, essential services, education or safety-sensitive products—can require complex documentation, testing and human oversight. If standards are incomplete and national authorities lack staff, forcing immediate compliance could produce paperwork without meaningful protection. Smaller firms, in particular, may struggle to interpret rules written with large technology companies in mind.
The counterargument, however, is stronger than the language of “breathing room” suggests. Regulation is not merely a technical checklist; it is a public signal about what kinds of failure society will tolerate. A system that wrongly rejects a job applicant, misidentifies a person or makes a dangerous recommendation can cause harm before a delayed obligation arrives. By then, affected people may have little practical route to challenge a decision, and the companies responsible may argue that the rules did not yet apply.
The problem is not only timing
Europe’s policymakers also face a competitiveness dilemma. The continent has often been accused of regulating products it does not build. Businesses warn that uncertainty will push investment elsewhere, while governments want domestic companies to scale quickly. That concern should not be dismissed as lobbying: predictable, workable rules are an economic asset, and compliance costs can fall hardest on smaller developers.
Yet speed and accountability are not opposites. Clear templates, shared testing infrastructure, regulatory sandboxes and well-funded national authorities could reduce costs without suspending the principle of oversight. The European Commission’s own proposal recognizes that implementation depends on supporting infrastructure and fixed dates, rather than indefinite postponement.[1] The policy question should therefore be how to make compliance feasible—not whether safety rules are inconvenient.
This is especially urgent because AI governance is becoming a contest between jurisdictions. China’s economic planning agency has outlined an AI governance agenda for the next five-year period, while Chinese technology researchers have publicly acknowledged that open-weight models carry inherent risks.[3] Europe cannot assume that restraint alone will make it a leader. But neither should it copy a race in which social costs are treated as evidence of progress.
A better bargain
Our view is that the EU should accept narrowly defined implementation relief only if it comes with enforceable safeguards. Delays should be tied to published technical standards, transparent progress reports and a firm sunset date. Companies deploying high-risk systems should still have to identify them, preserve records, inform affected people and provide meaningful human review. Regulators should publish anonymized enforcement data so the public can see whether the system is working.
Industry deserves certainty. Citizens deserve protection before, not after, an AI failure becomes a scandal. The Union can provide both by simplifying compliance while keeping accountability intact. If Europe postpones the rules, it must not postpone the duties that make those rules credible.