Elon Musk’s artificial intelligence company is facing intense backlash after reports revealed that its new video generator, Grok Imagine, produced sexually explicit clips of Taylor Swift without users asking for such content.

Critics and online safety experts are calling the case a disturbing example of the dangers of unchecked AI technology.

According to The Verge, Grok Imagine’s controversial “spicy” mode was tested by journalists to examine the platform’s guardrails. What they found was alarming—when prompted with a harmless request for “Taylor Swift celebrating Coachella with the boys,” the AI generated still images of the pop star wearing a dress alongside a group of men. But with “spicy” mode activated, the software automatically altered the scenario to show Swift ripping off her dress and dancing topless, in scenes described as “completely uncensored” and “completely exposed.”

Jess Weatherbed, the journalist behind the test, told the BBC that she “in no way asked it to remove her clothing” and was shocked at how quickly the AI turned an innocent prompt into explicit deepfake content. Similar results were reported by Gizmodo, which found that searches for other famous women also returned explicit material, though sometimes blurred or marked as “video moderated.”

Criticism from experts and lawmakers

Law professor Clare McGlynn, a leading voice in online abuse prevention, said the case shows this is “not misogyny by accident, it is by design.” She accused platforms like X (formerly Twitter) of making “a deliberate choice” not to implement stronger protections, despite knowing the risks.

Deepfakes—AI-generated images or videos that replace one person’s face with another—have been a growing online threat, especially for women in the public eye. In January 2024, pornographic deepfakes of Taylor Swift went viral on X and Telegram, gaining millions of views before X temporarily blocked searches for her name.

Perhaps more troubling, Grok Imagine reportedly failed to meet the UK’s newly introduced age verification laws for explicit content. Weatherbed signed up for the £30 paid version using a brand new Apple account. While the system asked for her date of birth, there were no additional verification checks—something UK law now requires to be “technically accurate, robust, reliable and fair.”

Under the Online Safety Act, which came into force at the end of July, platforms with generative AI tools capable of producing pornographic material must have proper safeguards in place. Ofcom, the UK’s media regulator, told the BBC it is “aware of the increasing and fast-developing risk GenAI tools may pose” and is working to ensure companies comply.

Push for stronger deepfake laws

Currently, creating pornographic deepfakes is illegal in the UK if they involve minors or are used for revenge porn. However, an amendment—drafted with the help of Prof McGlynn—will soon make it illegal to generate or request any non-consensual pornographic deepfake. Baroness Owen, who proposed the amendment in the House of Lords, said, “Every woman should have the right to choose who owns intimate images of her… whether she be a celebrity or not.”

The Ministry of Justice echoed that stance, calling sexually explicit deepfakes “degrading and harmful” and vowing to “ban their creation as quickly as possible.”