EU Bans Non-Consensual Intimate Deepfakes After Grok Scandal

On March 11, 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 using artificial intelligence — a decision that became a direct consequence of the scandal surrounding Grok.

What Scandal Prompted the Ban?

The amendment is directly linked to the public outcry surrounding Grok — an AI product found at the center of a scandal related to creating non-consensual intimate images using generative AI. This story became the catalyst that brought the issue to the level of legislative amendments to the AI Act — a clear illustration of the specific harm that image generation tools without built-in restrictions are capable of causing if they can be used to create sexualized content of real people without their consent.

Grok is a product of the company xAI, integrated among other things into image generation functionality available to a broad audience of users. It is precisely the availability of such tools to a mass audience, rather than only to specialized actors, that makes the problem of non-consensual intimate deepfake images systemic: virtually any person whose photographs are publicly available can potentially be harmed, not just public figures who have traditionally been the focus of earlier regulation on defamation and image protection.

• Date of political agreement: March 11, 2026
• Regulation: amendments to the EU's AI Act
• Essence of the amendment: direct ban on non-consensual intimate images created by AI
• Reason: scandal surrounding Grok
• Coalition of 57 European Parliament members was required for passage

How the Decision Was Made

According to The Next Web, assembling a coalition of 57 European Parliament members was needed to move the amendment to political agreement — a decision that resulted from targeted legislative pressure, rather than an automatic extension of already existing AI Act norms. Even within the framework of already adopted and considered advanced AI regulation, specific and detailed bans appear reactively — as a response to a resonant incident, not as a predetermined measure.

The mere fact that a targeted amendment to an already existing law required mobilizing dozens of deputies and a public scandal illustrates how complicated the legislative process at EU level is even regarding relatively unambiguously harmful practices. This also points to a general pattern in AI regulation: basic framework norms are adopted in advance, while specific, detailed bans on individual types of harm most often appear after the fact, as a reaction to an already occurred resonant incident, rather than as a result of preventive risk analysis.

What Will Change for AI Platforms?

The AI Act even before this amendment was considered the most comprehensive normative act on artificial intelligence among major jurisdictions worldwide, introducing requirements for transparency, risk assessment, and restrictions on high-risk systems. Adding a separate, explicitly formulated ban on non-consensual intimate deepfake images moves this category of harm from the category of "potentially falls under general norms on harmful content" into the category of directly prohibited practice — which for platforms providing image generation tools in the EU means the necessity of embedding appropriate filters and restrictions under threat of direct violation of law, rather than a general and more vague norm.

For global AI companies operating in the European market, such targeted amendments mean a constantly growing and ever more detailed list of product requirements for image and video generation: it is insufficient to bring a system into compliance with basic AI Act principles once; one must regularly track new precedents and amendments responding to specific types of harm that manifest only after the product has reached the market and become widely used.

For users, the very fact that such a ban had to be introduced as a separate amendment, rather than being initially included in the basic version of the AI Act, serves as a reminder: regulation almost always lags behind technology rather than preceding it, and until an explicit ban on a specific harmful practice appears, that practice can remain formally not prohibited long enough to cause real harm to specific people.