Grok Under Fire: Non-Consensual Sexual Deepfakes Persist

Dozens of non-consensual sexual deepfakes involving celebrities and U.S. Representative Alexandria Ocasio-Cortez remain accessible on Grok.com, according to Wired. Despite the restrictions announced by xAI last January, public workarounds are circulating, allowing the generation of this explicit content.

Grok Imagine, xAI's tool, hosts photorealistic images of women in simulated sexual situations. At the end of December 2025, Elon Musk announced that Grok could edit images from X.com. Within eleven days, the Center for Combating Digital Hate reported approximately 3 million sexualized images, of which 23,000 involved minors, according to a letter from 35 U.S. attorneys general.

Wired submitted two prompts extracted from Grok to other platforms. ChatGPT, Meta AI, and Claude rejected them as inappropriate, while Google Gemini validated one out of two. By searching updated specialized forums, any user can learn to artistically reframe their queries or generate neutral images beforehand to bypass the filters.

For Henry Ajder, a deepfake specialist, xAI and X are lagging behind the safeguards of other mainstream tools. In its preliminary findings published this week, the Office of the Privacy Commissioner of Canada concluded that xAI's measures lack evidence of effectiveness.

In January, xAI restricted the modification of images of real people in revealing outfits to paying subscribers on X. However, on the standalone Grok app, these restrictions appear ineffective. By early 2026, SpaceX absorbed xAI in a merger valued at approximately €1.15 trillion. SpaceX estimated the legal risks associated with Grok at over €460 million in its filing with the SEC. The overall litigation budget for SpaceX exceeds $500 million, with part of that amount covering Grok-related disputes. SpaceX classified Grok as one of the active sources of litigation related to non-consensual deepfakes and described these litigations as factors that could expose the company to sanctions for potentially explicit content and non-consensual or exploitative images.

The TAKE IT DOWN Act, a federal law enacted in May 2025 and fully in effect since May 19, 2026, requires platforms to remove non-consensual intimate images within 48 hours of a report. On March 26, the Amsterdam District Court ordered xAI and X to cease any distribution of non-consensual sexual images in the Netherlands, under penalty of €100,000 per day, capped at €10 million. To justify this decision, the judge cited a test conducted by Offlimits on March 9, which produced a sexualized video of a real person from a single photo, without consent verification.

Baltimore was the first U.S. city to sue xAI, followed by a class action from three teenage girls in Tennessee. Ashley St. Clair, the mother of one of Elon Musk's children, initiated a separate legal action after deepfakes of her were circulated on X. In June 2026, British MP Jess Asato filed a complaint in the High Court.

After the publication of the investigation, xAI removed the incriminating content, and X deleted the relevant links, without responding to requests for comment.