EU to ban non-consensual intimate deepfakes after the Grok scandal

On 11 March 2026, the European Parliament reached a political agreement on amendments to the European Union's AI Act that, for the first time, introduce a direct ban on creating non-consensual intimate images with artificial intelligence. The decision was a direct consequence of the scandal around Grok.

What scandal triggered the ban?

The amendment is directly tied to the public outcry over Grok, an AI product at the center of a scandal over the creation of non-consensual intimate images with generative AI. That story was the catalyst that pushed the issue to the level of legislative amendments to the AI Act. It illustrates the concrete harm that image-generation tools without built-in restrictions can cause when they can be used to create sexualized content about real people without their consent.

Grok is a product of xAI, integrated with, among other things, image-generation functionality available to a broad base of users. It is the availability of such tools to a mass audience, and not only to specialized actors, that makes the problem of non-consensual intimate deepfakes systemic: virtually anyone whose photos are publicly available can be affected, not just the public figures on whom earlier defamation and image-protection rules traditionally focused.

• Date of political agreement: 11 March 2026
• Regulation: amendments to the EU AI Act
• Substance of the amendment: a direct ban on AI-generated non-consensual intimate images
• Trigger: the Grok scandal
• Adoption required a coalition of 57 members of the European Parliament

How the decision was made

According to The Next Web, getting the amendment to a political agreement required a coalition of 57 members of the European Parliament. The outcome resulted from deliberate legislative pressure, not from an automatic extension of existing AI Act provisions. Even under AI regulation that is already adopted and considered advanced, specific bans appear reactively, as a response to a high-profile incident rather than as a measure planned in advance.

The very fact that a targeted amendment to a law already in force needed dozens of lawmakers to mobilize, along with a public scandal, shows how complicated the EU legislative process is, even for relatively clear-cut harmful practices. It also points to a general pattern in AI regulation: baseline framework rules are adopted in advance, while specific, detailed bans on particular harms most often arrive after the fact, in reaction to an incident that has already caused an outcry, rather than as the product of preventive risk analysis.

What will change for AI platforms?

Even before this amendment, the AI Act was considered the most comprehensive AI legislation among the world's major jurisdictions, introducing requirements for transparency, risk assessment and restrictions on high-risk systems. Adding a separate, explicitly worded ban on non-consensual intimate deepfakes moves this category of harm from "potentially covered by general rules on harmful content" to a directly prohibited practice. For platforms offering image-generation tools in the EU, that means building in the relevant filters and restrictions, under the threat of a direct violation of the law rather than of a general and vaguer norm.

For global AI companies operating in the European market, such targeted amendments mean a steadily growing and ever more detailed list of requirements for image and video generation products. It is not enough to align a system with the AI Act's basic principles once; companies need to track new precedents and amendments that respond to specific harms, which emerge only after a product has reached the market and come into mass use.

For users, the fact that this ban had to be introduced through a separate amendment, rather than being part of the original AI Act, is a reminder that regulation almost always catches up with technology rather than getting ahead of it. Until an explicit ban appears, a harmful practice can remain formally not prohibited for long enough to cause real harm to real people.