xAI's Grok Imagine 'spicy mode' used to make nonconsensual Taylor Swift deepfakes
A Verge reporter got explicit video of the singer on a first, unjailbroken attempt, unlike rival tools from Google and OpenAI which blocked celebrity nudity outright.
- Culture & impact
- Security & misuse
- Safety & alignment
- Major
Journalist Jess Weatherbed of The Verge reported that xAI’s newly launched Grok Imagine video generator produced sexually explicit deepfake video of Taylor Swift using its “spicy” preset, without her requesting nudity or attempting to bypass any filter. Asked for an image of Swift “celebrating Coachella with the boys,” the tool generated dozens of images in suggestive outfits; switching on the spicy preset and animating one produced, on what was reported as the first attempt, a video of a Swift lookalike stripping and dancing in a thong for a computer-generated crowd.
Grok Imagine, released in early August 2025 as part of the Grok app, offered four video presets — “Custom,” “Normal,” “Fun” and “Spicy” — the last intended to loosen content restrictions relative to the others. Reporting on the episode noted that competing video generators, including Google’s Veo and OpenAI’s Sora, blocked attempts to sexualise images of real people at the time, making xAI’s tool an outlier among major labs on this specific failure mode. X’s safety team said it was removing the identified images and acting against accounts that had posted them, and Grok’s own account, replying to users, acknowledged that the spicy preset had generated the content, attributing the behaviour to xAI’s general design preference for fewer restrictions than competitors impose.
The episode drew attention partly because it echoed an earlier, unrelated case: sexual deepfakes of Swift, made with different tools, had spread widely on X in January 2024 before the platform intervened. This time the images came from a tool xAI itself operated and had just shipped. It proved to be an early instance of a problem that recurred and expanded through Grok Imagine over the following months, as reporting through late 2025 and into 2026 continued to document sexualised imagery, including images alleged to depict minors, generated by the same feature.