EU Parliament adopts ban on sexualized deepfake AI and tightens AI Act

What is new since June 16, 2026

At the core of the amendment is the extension of Article 5 of the AI Regulation. In future, the placing on the market, putting into service, or use of an AI system that generates or manipulates realistic image, video, audio, or similar content in which intimate body parts of an identifiable natural person are depicted or in which an identifiable person involved in explicitly sexual acts is portrayed is prohibited, unless the person concerned has given freely given, specific, informed, unambiguous, and explicit consent. Systems in which the visibility of intimate body parts is neither increased nor the nature of the sexual acts altered are exempt.

The regulation on AI deepfakes is therefore to enter into force on August 2, 2026. Most of the other provisions of the AI Act will also apply from August 2, 2026. For standalone high-risk AI systems, however, the requirements will only apply from December 2027. For AI systems that are embedded as safety components and fall under sectoral EU legislation on safety and market surveillance, the deadline is August 2028.

Entry into force and deadlines

Also new is the mandatory labeling of AI-generated content: watermarks become compulsory. The corresponding application for labeling AI-generated content likewise enters into force on August 2, 2026. This is intended to make it recognizable when image, video, or audio content has been generated or altered by machine – an aspect that has gained importance particularly during election campaigns and disinformation campaigns.

In the negotiations, the definition of safety components in AI systems was particularly controversial. According to Recital 7, the previous definition did not provide the necessary precision for providers of AI systems to determine whether an AI system is to be classified as a safety component. In future, its intended purpose must consist in averting risks to the health and safety of persons or property. This is intended to clarify the distinction between conventional safety components and standalone AI applications.

Criticism of the expansion of the high-risk concept

Critics, however, warn of a creeping expansion of the high-risk concept. The new regulation carries the risk of extending the classification as a high-risk AI system beyond what is justified by the risk-based approach of this regulation. Business associations fear additional bureaucracy and longer certification procedures, while data protection organizations and women's rights groups expressly welcome the step.

The amendment is part of a larger reform package that adapts the AI Regulation, which only entered into force in 2024, to the rapid developments in the field of generative models. The EU Commission had already proposed the revision last year in order to respond to the emergence of increasingly powerful image and language models.

With the adoption in plenary, the legislative process at European level is not yet concluded. The EU member states must now find a common position in the Council before the regulation can enter into force. The newly created European AI Office is then to monitor compliance with the provisions and act as the central supervisory authority.

In parallel, the member states are working on national accompanying laws that regulate, among other things, penalties for violations of the deepfake ban. Observers in Strasbourg expect that the first fines will be imposed no earlier than 2027, when the high-risk provisions take effect. Until then, it remains to be seen how the requirements will affect the business models of large platforms, which in the past had only hesitantly removed numerous cases of non-consensual intimate content.