Commentary. Europe’s attempt to regulate artificial intelligence is entering the phase that matters most: implementation. The European Union once presented its AI Act as proof that democratic governments could set enforceable limits on powerful technology. Now, delays and proposed changes risk turning that promise into another long timetable in which companies move faster than the institutions meant to oversee them.
That does not mean every criticism of the law is misplaced. Businesses are right that compliance can be expensive, particularly for smaller companies that lack specialist legal and technical teams. Developers also argue that unclear standards can discourage investment, push research abroad and make it harder to deploy useful systems in health care, education and public administration. Those concerns deserve more than slogans.
But the answer cannot be to treat delay as neutrality. A postponed safeguard is still a missing safeguard when automated systems influence employment, credit, public benefits, policing or medical decisions. The European Commission’s AI Omnibus has extended the timeline for rules covering high-risk systems to December 2027, according to reporting by CNN. That is a significant amount of time in an industry whose products and capabilities can change within months.[1]
The case for a slower approach
The strongest argument for flexibility is practical. Regulators cannot write intelligent rules without understanding the systems they supervise, and companies cannot comply with obligations that remain technically vague. A rushed regime could create paperwork without improving safety, or freeze older assumptions into law while models evolve.
There is also a legitimate concern about fragmentation. National regulators, American states and other governments are developing different approaches. A company operating across borders may face overlapping audits, disclosure requirements and liability rules. Policymakers who ignore that burden could protect incumbents while making it harder for smaller competitors to enter the market.
Effective regulation should be predictable enough to follow and adaptable enough to remain relevant.
That principle supports phased implementation, technical guidance and assistance for smaller firms. It does not support indefinite postponement.
Why accountability cannot wait
The costs of weak oversight are not theoretical. AI systems can reproduce discriminatory patterns, expose private information, generate convincing falsehoods and fail unpredictably in high-stakes settings. Transparency alone is not enough if affected people cannot challenge a decision or obtain meaningful human review.
Recent developments show the regulatory gap widening rather than closing. A policy briefing from law firm Faegre Drinker said the United States had rejected international coordination on AI governance while several US states advanced their own oversight measures, including new arrangements for AI auditors in California.[2] That patchwork may encourage experimentation, but it also leaves basic protections dependent on geography.
Europe should not respond by abandoning ambition. It should respond by making the rules usable: publish clear technical standards, create a common registry of high-risk systems, fund independent testing and require companies to document how consequential systems are trained, evaluated and monitored after deployment.
What responsible flexibility looks like
A sensible compromise would distinguish between genuinely low-risk applications and systems that can materially affect a person’s rights or livelihood. Harmless productivity tools should not face the same process as software used to determine access to housing or treatment. At the same time, companies should not be allowed to evade safeguards simply by describing a consequential system as experimental.
Enforcement should also be proportionate. Regulators can begin with warnings and deadlines for correction, while imposing serious penalties for concealment, repeated violations or deliberate deployment of unsafe systems. Independent researchers and journalists need protected access to evidence of harm, because regulators cannot detect every failure alone.
Our view is straightforward: Europe should welcome targeted simplification, but reject delay without a public safety rationale. The credibility of the AI Act will not be measured by the elegance of its text. It will be measured by whether an ordinary person can contest a harmful automated decision, and whether a company knows that preventable damage will carry consequences.
Innovation needs room to grow. It also needs trust. The fastest route to durable public confidence is not weaker rules, but rules that are clear, enforceable and applied before the next crisis makes their absence impossible to ignore.