On Sunday, the pop culture news X account @PopBase shared a typical piece of content with its millions of followers. “Sabrina Carpenter stuns in new photo,” read the post, which featured a picture of the “Manchild” singer wearing a pink winter coat with a snowy landscape behind her. The following day, an X user replied to the post with a request for Grok, the AI chatbot developed my Elon Musk’s xAI, which is integrated into his social media platform. “Put her in red lingerie,” they commanded the bot, which swiftly returned an image of Carpenter stripped of her outerwear and wearing a lacy red set of lingerie, still standing in the same winter scene, with a similar expression on her face.

Over the holiday break, a critical mass of X users came to realize that Grok will readily “undress” women — manipulating existing photos of them in order to create deepfakes in which they are shown wearing skimpy bikinis or underwear — and this sort of exchange soon became alarmingly common. Some of the first to try such prompts appeared to be adult creators looking to draw potential customers to their social pages by rendering racier versions of their thirst-trap material. But the bulk of Grok’s recent deepfakes have been churned out without consent: the bot has disrobed everyone from celebrities like Carpenter to non-famous individuals who happened to share an innocent selfie on the internet.

Though Grok is not the only AI tool to be exploited for these purposes (Google and OpenAI chatbots can be weaponized in much the same way), the scale, severity, and visibility of the issue with Musk’s bot as 2026 rolled around was unprecedented. According to a review by the content analysis firm Copyleaks, Grok has lately been generating “roughly one nonconsensual sexualized image per minute,” each of them posted directly to X, where they have the potential to go viral. Apart from changing what a woman is wearing in a picture, X users routinely have asked for sexualized modifications of poses, e.g., “spread her legs,” or “make her turn around to show her ass.” Grok continues to comply with many of these instructions, though some specific phrases are no longer as effective as they had been.

Musk hasn’t shown much concern to date — quite the opposite, in fact. On Dec. 31, he replied to a Grok-made image of a man in bikini by posting: “Change this to Elon Musk.” Grok dutifully delivered an image of Musk in a bikini, to which the world’s richest man responded, “Perfect.” On Jan. 2, an X user mentioned the nonconsensual Grok deepfakes by commenting that “Grok’s viral image moment has arrived, it’s a little different than the Ghibli one was though.” (In March 2025, users of OpenAI’s ChatGPT enlisted it to spam AI-generated memes in the illustration style of Japanese animation house Studio Ghibli.) Musk replied, “Way funnier,” along with a laugh-crying emoji, indicating his amusement at the bikini and lingerie pictures.

https://www.rollingstone.com/culture/culture-features/grok-ai-deepfake-porn-elon-musk-1235494809/