Strasbourg: The European Parliament has done two things at once that sit awkwardly together. On 16 June 2026 it gave final approval to the Digital Omnibus on AI, a package of targeted amendments to the AI Act, and the Council is expected to formalise the text on 29 June before publication in the Official Journal. The package draws a new red line by banning AI systems that generate non-consensual intimate imagery and child sexual abuse material. In the same breath, it postpones several of the obligations that were meant to police high-risk systems. The pairing reveals how the Union now reasons about technology regulation, and why the result satisfies almost no one fully.
The prohibition itself is unambiguous and overdue. So-called nudifier tools, which fabricate explicit images of real people without consent, have proliferated faster than any enforcement mechanism could respond. The new rule bans both placing such systems on the EU market and deploying them for those purposes, with compliance required by 2 December 2026. By treating non-consensual intimate imagery and CSAM as prohibited practices rather than merely high-risk ones, lawmakers move them into the AI Act’s most stringent tier, where the question is not how to manage a system but whether it may exist at all. For a body often accused of regulating process over outcome, this is a clear substantive judgement about dignity and consent.
The harder questions surround what the omnibus delays. The same legislative vehicle pushes back compliance deadlines for a range of high-risk AI obligations, the rules governing systems used in hiring, credit scoring, education and access to essential services. Supporters frame the postponement as pragmatism. Standards bodies have not finished the technical specifications that companies need to comply, and forcing obligations to bite before the supporting guidance exists would have produced confusion rather than protection. Industry has argued for breathing room, and the Commission’s broader simplification agenda has been sympathetic to the complaint that the original timeline was unrealistic.
Critics see something less benign. Civil-society groups and several legal scholars warn that delaying high-risk oversight lets exactly the systems most likely to affect people’s livelihoods operate longer without the safeguards the Act promised. The worry is structural: a regime that bans the most viscerally objectionable uses while loosening the timetable for the most consequential everyday ones may optimise for political legibility rather than for aggregate protection. Nudifiers are easy to condemn. An opaque algorithm quietly filtering job applicants is harder to dramatise but touches far more lives.
That tension is the real story of the omnibus. The AI Act was sold as a risk-based pyramid, with prohibited practices at the apex and graduated duties below. The June amendments sharpen the apex and blur the middle. Whether that is coherence or contradiction depends on what one thinks the Act is for. If the goal is to prevent categorical harms to fundamental rights, banning synthetic sexual abuse imagery is precisely the right use of the prohibition tier. If the goal is to build everyday accountability into the systems that allocate opportunity, the delay is a retreat dressed as housekeeping.
There is also a credibility dimension that extends beyond Europe’s borders. The EU has staked its regulatory authority on being the jurisdiction that writes the rulebook others adopt by default. Each postponement invites the argument that the framework was overambitious, and each invites trading partners and domestic lobbies to press for the next extension. The Commission will counter that adjusting deadlines to technical reality strengthens the Act by making it enforceable. Both claims can be true, which is why the coming year of implementation matters more than the vote itself.
For now, the practical takeaways are concrete. Developers of intimate-imagery tools face an outright ban with a firm deadline. Providers of high-risk systems gain time but not absolution, since the obligations are deferred rather than removed. And the EU has signalled that its appetite for prohibition remains strongest where harms are visceral and consensus is easy, and weakest where the costs of compliance fall on established industry. That is a defensible political settlement. Whether it is the one that best protects fundamental rights is the question the next phase will answer.




