The EU Parliament voted this week to amend the Artificial Intelligence Act, approving a package that delays compliance deadlines for high-risk AI systems while introducing something genuinely new: an outright ban on so-called “nudifier” apps.

The votes were lopsided — 569 in favour, 45 against, 23 abstentions. The direction of travel was never in doubt.

What’s Being Delayed

The EU’s original AI Act compliance timelines were ambitious to the point of impracticality. The new “omnibus” simplification package pushes the application dates for high-risk AI systems — those used in biometrics, critical infrastructure, education, employment, law enforcement, and border management — to 2 December 2027. For AI systems covered by existing sector-specific EU product legislation (medical devices, toys, radio equipment), the deadline moves to 2 August 2028.

The Parliament also pushed the watermarking deadline to 2 November 2026, giving providers roughly 19 months to implement systems that label AI-generated audio, image, video, and text content.

The official reason: guidance and standards needed for companies to actually comply weren’t ready. That’s a reasonable argument — you can’t enforce compliance with standards that don’t exist. But it also illustrates the predictable failure mode of sweeping, complex legislation: by the time you’ve written the rules, the technology and the practical guidance for following them both need updating.

The Nudifier Ban

The more notable development is the ban on nudifier apps — AI systems that create or manipulate sexually explicit images of real people without their consent. This isn’t a延迟 or a simplification. It’s a new prohibition that didn’t exist in the original Act.

The ban won’t apply to AI systems with “effective safety measures” preventing the creation of such images. That framing is interesting: it means the obligation is on the system to prevent abuse, not on bad actors to refrain from abusing it. Whether that works in practice depends entirely on whether the safety measures are technically robust.

The Substantive Issue

What this vote actually demonstrates is that the EU is quietly retreating from the AI Act’s original ambition. The delays are substantial — two years for high-risk systems. The Omnibus VII package is framed as simplification, but simplification usually means weakening in practice, because the pressure to simplify comes from those who find the rules burdensome.

The EU’s AI Act was supposed to be the world’s most comprehensive AI regulatory framework. What it’s becoming is something more like a promise that gets renewed every two years when the deadlines approach.

Sources: European Parliament press release, EU Council press release