AI-Generated Explicit Images of Public Figures: What the Oversight Board Decided and Why It Matters

Advances in generative AI have made it easier than ever to create realistic explicit images of real people without their consent. The Oversight Board of Meta recently examined two such cases, involving female public figures from India and the United States. Their decision offers important lessons on how social platforms should protect people from AI-driven sexual abuse while keeping their rules clear and enforceable.

The first case involved an Instagram account dedicated to AI-generated or AI-manipulated images of Indian women. It posted an image of a nude woman, seen from behind with her face visible, that resembled a well-known Indian public figure. A similar image showing the same woman in a swimsuit appeared in the same series, likely the original photo altered by AI. The second case involved an AI-generated image shared in a Facebook group dedicated to AI creations. It showed a nude woman being groped, styled to resemble a named American public figure mentioned in the caption.

In the Indian case, a user reported the post for pornography, but the report was automatically closed because it was not reviewed within 48 hours. The user's appeal was closed the same way. After the Board selected the case, Meta concluded its original decision was wrong, removed the post under its bullying and harassment standard, disabled the account, and added the image to a Media Matching Service bank. In the American case, the image had already been added to a matching bank after violating Meta's policy, so it was removed automatically. The user's appeal was also automatically closed, prompting an appeal to the Board.

Deepfake intimate images use computer-manipulated media to depict real people in sexualized ways. Creating them requires fewer source images than before, and online deepfake videos reportedly rose by 550 percent since 2019. Research cited in the case found that the vast majority of deepfake content is sexual and overwhelmingly targets women. Female politicians, journalists, human rights defenders, and celebrities are frequently targeted across many countries, and cases involving children and teenagers have also been documented. In conservative communities, such content can lead to severe real-world consequences, including violence.

Research on image-based sexual abuse shows that victims often experience shame, helplessness, anxiety, isolation, humiliation, and distress. They may also lose their sense of dignity, safety, and self-confidence. Researchers have argued that sexualized deepfakes can cause harm comparable to that of real non-consensual intimate images. Because the damage comes from the creation, sharing, and viewing of these images, labeling them as manipulated does not adequately address the harm.

The Board concluded that both images violated Meta's prohibition on manipulated sexualized degrading images. The visual changes were clear, and contextual cues such as hashtags and the posting environment showed AI generation. The American image also violated Meta's rules on adult nudity and sexual activity because it depicted sexual touching. The Board found that removing both posts was consistent with Meta's human rights responsibilities. It also noted that deepfake intimate images require removal because the nature and severity of the harm cannot be resolved by a warning label alone.

The Board found that the phrase "manipulated sexualized degrading images" was not clear enough for ordinary users, even if the users in these cases should have understood it. Meta explained the term to mean images sexualizing people in ways they would likely not have wanted and that are therefore perceived as degrading. The Board said "non-consensual" more clearly describes the core issue. It also found that the word "manipulated" is too narrow for modern techniques, including generative AI, and should be broadened to cover all forms of media manipulation.

The Board recommended that Meta move the prohibition on manipulated sexualized degrading images into its Adult Sexual Exploitation Community Standard rather than the bullying and harassment policy. Users are unlikely to see deepfake intimate images as bullying, since research shows they are often shared to attract audiences, monetize pages, or drive traffic to other sites, including pornographic ones. Meta should also replace "degrading" with "non-consensual," replace "manipulated images" with a broader term for manipulated media, and consider renaming the policy to "non-consensual sexual content." Finally, AI-generated or manipulated context showing nudity or sexualization should be treated as a signal of non-consent, and such content should not need to be non-commercial or privately produced to violate the rule.

The Board expressed concern that appeals about image-based sexual abuse were closed automatically. Waiting 48 hours for review can seriously harm victims. Although the Board does not yet have full information on how widely Meta uses automatic closure, it considers the issue to have significant human rights implications and called for a risk assessment and corrective action. Meta also only added the Indian public figure's image to a matching bank after the Board asked questions, and the Board noted that victims of deepfake intimate images often remain unknown to the public, leaving them to find and report content themselves.

This decision highlights that AI-generated sexual imagery is a serious form of abuse that demands fast, decisive removal rather than weak labeling. Platforms must write clear rules, place them in the right policy categories, respond quickly to appeals, and proactively protect vulnerable users, especially women and girls. As governments develop new laws on deepfakes, social media companies must remain the first line of defense to prevent harm and protect the dignity and safety of people affected by synthetic sexual content.