EU negotiators add explicit ban on non‑consensual 'nudifier' AI to AI Act

The EU’s AI Act just got a very specific new red line. Negotiators from the European Parliament and Council agreed on May 7 to explicitly ban AI systems used to generate child sexual abuse material and non-consensual intimate content — the category people usually mean when they say “nudifier” apps. But the same deal also softens other parts of the law. Basically, Brussels is getting tougher on one ugly consumer abuse case while giving companies more time and more clarity elsewhere. ### What exactly did they ban? The new prohibition targets AI systems that create sexual or intimate content involving an identifiable person without that person’s consent, plus systems that generate child sexual abuse material. The rule is broader than just fake nude photos. It covers images, video, and audio. And it does not only hit end users. It also covers providers that put these systems on the EU market for that purpose, or that ship them without reasonable safeguards to stop that use. (consilium.europa.eu) ### Why does that matter so much? Because this was a glaring hole. The original AI Act had broad bans on a few unacceptable practices, but it did not cleanly single out the fast-growing market for sexualized deepfake tools. That left regulators trying to patch harms through privacy law, platform rules, or criminal law after the fact. This amendment moves the line upstream — closer to the tool itself. (europarl.europa.eu) ### Who has to change first? The nudifier ban moves fast. Companies will have until December 2, 2026 to bring systems into line. That is much sooner than the delayed deadlines for parts of the high-risk regime, which now land on December 2, 2027 for stand-alone high-risk systems and August 2, 2028 for AI embedded in regulated products. So the political signal is pretty clear — sexual exploitation risks get accelerated treatment, even while business-facing compliance gets more runway. (consilium.europa.eu) ### What got easier for companies? The biggest relief is around overlap. The deal says some AI used in machinery and other regulated products should not face duplicative compliance under both the AI Act and sector-specific product safety rules. It also narrows the definition of “safety component,” so AI that merely assists a user or optimizes performance does not automatically become high-risk if failure would not create a health or safety risk. (europarl.europa.eu) That matters because “high-risk” is the label that triggers the heavy obligations. ### Why were deadlines pushed back? Because the original timetable was colliding with reality. High-risk AI obligations were due to start on August 2, 2026, but companies and regulators were still waiting on standards, guidance, and practical tools. The Commission had proposed a more conditional delay tied to standards readiness. Negotiators scrapped that moving target and replaced it with fixed dates, which is simpler even if it slows enforcement. (europarl.europa.eu) ### What else changed under the hood? The agreement also reinforces the AI Office and tries to reduce fragmented governance. It keeps the AI Act’s basic risk-based structure, but tweaks how bias detection can work and restores some registration duties even for providers claiming an exemption from high-risk classification. In plain English — fewer overlapping rules, but not a free pass. (consilium.europa.eu) ### Is this final? Not yet. This is a provisional political agreement, so Parliament and Council still need to formally adopt the text. But the core shape now looks set: one new outright ban for a highly visible abuse category, paired with slower and more targeted rollout for the harder industrial parts of the AI Act. (consilium.europa.eu) ### Bottom line The EU did not rewrite the AI Act from scratch. It did something more tactical. It picked one harm that had become impossible to ignore — AI-generated sexual abuse and humiliation — and banned it explicitly, while making the rest of the regime more survivable for companies building industrial and safety-related systems. (consilium.europa.eu)