EU's AI Act: stepping back without stepping down
The EU is quietly rewriting the timeline of its landmark AI law, under pressure from U.S. tech giants and Washington. Pragmatic fix — or first retreat?
At a Glance:
The EU has pushed back compliance deadlines for high-risk AI systems to December 2, 2027, and to August 2, 2028, for manufacturers embedding AI in physical products.
U.S. tech platforms lobbied hard against the law’s timeline, securing the backing of the Trump administration in their push for concessions from Brussels.
The only genuinely new measure: a formal ban on software that generates non-consensual intimate imagery and child sexual abuse material.
The AI Act, a landmark law caught between ambition and industry reality
What was once heralded as the world’s first comprehensive, legally binding framework for artificial intelligence is now getting its first major overhaul — before most of its provisions have even taken effect. The European Commission, the EU’s executive arm, confirmed in early May 2026 a significant rescheduling of the AI Act’s rollout. Developers of so-called high-risk AI systems — those capable of affecting fundamental rights, such as biometric surveillance tools or automated hiring algorithms — now have until December 2, 2027, to achieve compliance. Manufacturers integrating AI into physical products, from children’s toys to elevators and medical devices, have been given an additional extension, to August 2, 2028.
Henna Virkkunen, the Finnish executive vice-president of the European Commission overseeing the EU’s technology sovereignty portfolio, framed the challenge in terms that reveal its core tension: companies and citizens, she noted, want two things from AI rules — the freedom to innovate, and the assurance that they are safe. Those two demands do not resolve themselves naturally. That unresolved tension is precisely what Brussels is now navigating in real time.
U.S. pressure: a decisive factor in Brussels’ retreat
This rescheduling did not emerge from internal deliberation alone. America’s major digital platforms — whose transparency, traceability, and risk-management obligations under the AI Act directly threaten their operating models in the EU market — organized effective lobbying campaigns, ultimately securing the public support of the Trump administration. Washington made clear to Brussels that an overly restrictive regulatory environment in Europe would constitute a systemic competitive disadvantage for U.S. operators.
This kind of diplomatic-commercial pressure is not without precedent. A similar dynamic played out, to varying degrees, during the negotiations over the GDPR — Europe’s sweeping data privacy law — as well as the Digital Markets Act, which regulates the behavior of major online platforms, and various EU digital tax initiatives. What is new here is the timing: the pressure is bearing down on a text still in its deployment phase, before any real-world impact assessment has been possible.
It is plausible that these delays represent a tactical concession by Brussels, designed to protect other dimensions of the transatlantic trade relationship at a moment when tariff negotiations are under acute strain. That hypothesis, unconfirmed at this stage, nonetheless warrants scrutiny.
What still holds: a hard line on non-consensual content
Amid the rollbacks and deadline extensions, one firm boundary has been drawn. The revised AI Act now explicitly bans software capable of generating non-consensual sexually explicit imagery — the so-called “nudification” applications — as well as any material constituting child sexual abuse. This provision was not present with this level of specificity in the original text.
For the average European user of an AI-powered service, this matters in concrete terms: until the broader AI Act provisions take effect, formal remedies for most algorithmic harms remain limited. But this ban is immediate and unconditional — one area where the law does not wait for industry to catch up.
The speed of its insertion is itself telling. Rather than emerging from a long legislative drafting process, this measure was fast-tracked in direct response to the proliferation of such content across social media platforms in recent months. This approach could be characterized as what legal scholars sometimes call reactive lawmaking — a response to visible harm rather than a proactive framework. That carries its own risks, including questions of proportionality and systemic coherence.
The bottom line
The AI Act is not dead — but it has entered a zone of uncertainty that reveals something deeper than a scheduling dispute. Each delay is presented as a pragmatic adjustment to avoid penalizing innovation. The real question, however, is not calendrical: it is whether Europe will retain, after successive postponements, the regulatory substance that made this law globally singular in the first place.
A pioneering framework hollowed out by deferred deadlines is, ultimately, policy without effect.
The tension between regulatory sovereignty and technological competitiveness will not be resolved by additional time. It requires a political answer that Brussels has yet to clearly articulate.
Sources: France Info · European Commission (press release IP_26_1024)