EU committee backs ban on deepfake-capable AI systems - with exceptions

The EU committees for the Internal Market and Civil Liberties approved the planned changes to the so-called Omnibus package on Wednesday. The aim is to regulate high-risk AI systems and also to ban certain deepfakes. This includes the AI generation of content depicting child abuse and intimate situations, as well as non-consensual sexual acts. It refers not only to the lack of permission for the act itself, but the lack of permission to artificially create such images.

The so-called Omnibus AI Act is part of the simplification agenda, where the EU is currently reviewing several existing laws for their currency and implementability. In this case, it is primarily about previously missing standards and tools for high-risk AI, which still need to be developed. Earlier this week, the European Council had already approved the changes to the Omnibus AI Act with a deadline extension for high-risk AI systems and the deepfake ban.

Now, the EU committees for the Internal Market and Civil Liberties have also overwhelmingly approved the proposed changes. According to an EU press release, 101 MEPs voted in favor and only 9 against, with 8 abstentions. The regulations also include that AI-generated content must be watermarked earlier than initially planned. Originally, AI providers were given a deadline of February 2, 2027 for this; now they are to implement it by November 2, 2026.

Ban on AI-generated deepfakes without consent

MEPs do not want to see so-called naked image generators at all. The trigger for this initiative was mass-produced images with very questionable content that people had created with Grok's image generator. Many posted the images on X. Grok is the image generator from Elon Musk's xAI, which also owns X. Users had instructed the AI to depict well-known personalities in provocative poses or during sexual acts, which Grok initially complied with.

Other AI systems refuse such instructions, which is why they are to be excluded from the planned ban. If these AI offerings have effective security measures against such image or video generation, the ban would not apply to them. However, it remains open whether and how AI providers are to prove this to the EU. It also remains unclear how AI systems can verify whether the depicted individuals have consented to the depiction or not.

Vote in the EU Parliament next week

Kim van Sparrentak, a Dutch environmental scientist and MEP for the Green Party, welcomes the law nonetheless. “This is a major success, especially for women and children in Europe,” she said, according to France 24. “Too many people have woken up one day in despair after finding naked photos of themselves and feeling violated, intimidated, and hunted.” Irish MEP Michael McNamara, who is involved in the AI legislation, added that the ban aims to eradicate “non-consensual naked image apps that have brought much suffering and profit to a few.”

After the two EU committees approved the changes to the Omnibus AI laws, the EU Parliament is expected to vote on them on March 26. If this plenary also agrees, negotiations with the EU Council will begin.

(fds)