Elon Musk’s xAI Tool Made Nude Deepfakes of Taylor Swift, Bypassing Its Own Rules

An AI video tool from Elon Musk’s xAI is producing nude deepfakes of celebrities—including Taylor Swift—with minimal safeguards, raising alarms over AI content moderation and non-consensual imagery.

The feature, called Grok Imagine, launched Tuesday in the Grok app for iOS and Android. It lets users animate AI-generated images into short videos under four style presets: Normal, Fun, Custom, and Spicy. Choosing Spicy can create nude or sexually suggestive clips;  according to The Verge, the tool automatically generated a video of a Swift lookalike undressing, unbidden by its reporter. After the Swift deepfakes became public, on Wednesday, xAI restricted the ability to generate spicy versions of the pop star.

Perhaps the earlier uncensored version of the video generator explained its immediate popularity: Musk said Imagine usage had grown more than 50%, calling it “Hyperviral,” with 20 million images generated on Tuesday alone.

Unlike platforms such as Google’s Veo 3 and OpenAI’s Sora—which include safeguards to block adult content and celebrity likenesses—Grok Imagine appeared to sidestep those protections, despite xAI’s own policies. The company’s terms state that users may not depict “likenesses of persons in a pornographic manner.”

Musk has not yet publicly addressed the controversy, which comes a year after AI deepfakes targeted Swift during the 2024 U.S. election season.

In August 2024, deepfake images appeared to show Swift endorsing then-candidate Donald Trump. The incident led to widespread condemnation and renewed calls for the regulation of generative AI. Swift later endorsed Trump’s opponent, former Vice President Kamala Harris.

In May, Congress passed the “Take It Down Act,” giving victims of nonconsensual intimate imagery a formal process to request takedowns from platforms. It remains unclear whether xAI will restrict Grok’s ability to generate explicit content featuring real people.

This is the latest kerfluffle involving xAI releasing tools that were not properly vetted. In July, Grok went on a racist tirade, calling itself “MechaHitler.” The chatbot began making antisemitic statements and claims of pervasive ideology, propaganda, and anti-white stereotypes in Hollywood movies.

The news of Grok’s ability to generate nude videos of Swift comes a day after a federal judge struck down a California law restricting AI-generated deepfakes during elections.