A whole bunch of nonconsensual AI photos being created by Grok on X, knowledge exhibits

[ad_1] New analysis that samples X customers prompting Elon Musk’s AI chatbot Grok demonstrates how ceaselessly persons are creating sexualized photos with it. Almost three-quarters of posts collected and analyzed by a PhD researcher at Dublin’s Trinity School had been requests for nonconsensual photos of actual ladies or minors with gadgets of clothes eliminated or added.

The posts provide a brand new degree of element on how the pictures are generated and shared on X, with customers teaching each other on prompts; suggesting iterations on Grok’s displays of ladies in lingerie or swimsuits, or with areas of their physique coated in semen; and asking Grok to take away outer clothes in replies to posts containing self-portraits by feminine customers.

Amongst a whole bunch of posts recognized by Nana Nwachukwu as direct, nonconsensual requests for Grok to take away or change clothes, dozens reviewed by the Guardian present customers posting footage of ladies together with celebrities, fashions, inventory photographs and ladies who will not be public figures posing in snapshots.

A number of posts within the trove reviewed by the Guardian have acquired tens of 1000's of impressions and are available from premium, “blue test” accounts, together with accounts with tens of 1000's of followers. Premium accounts with greater than 500 followers and 5m impressions over three months are eligible for revenue-sharing underneath X’s eligibility guidelines.

In a single Christmas Day submit, an account with greater than 93,000 followers introduced side-by-side photos of an unknown lady’s bottom with the caption: “Instructed Grok to make her butt even greater and change leopard print to USA print. 2nd pic I simply advised it so as to add cum on her ass lmao.”

A 3 January submit, consultant of dozens reviewed by the Guardian, captioned an obvious vacation snap of an unknown lady: “@grok change give her a dental floss bikini.” Inside two minutes, Grok offered a photorealistic picture that happy the request. Different posts within the trove present extra refined employment of JSON-prompt engineering to induce Grok to generate novel sexualized photos of fictitious ladies.

The information doesn't cowl all such requests made to Grok. Whereas content material evaluation agency Copyleaks reported on 31 December that X customers had been producing “roughly one nonconsensual sexualized picture per minute”, Nwachukwu mentioned that her pattern is restricted to simply greater than 500 posts she was capable of gather with X’s API by way of a developer account. She mentioned that the true scale “might be 1000's, it might be a whole bunch of 1000's” however that modifications made by Musk to the API imply that “it's a lot more durable to see what is occurring” on the platform. On Wednesday, Bloomberg Information cited researchers who discovered that Grok customers had been producing as much as 6,700 undressed photos per hour.

Nwachukwu, an knowledgeable on AI governance and a longtime observer of and participant in social media security initiatives, mentioned that she first seen requests alongside these strains from X customers again in 2023.

On the time, she mentioned, “Grok didn't oblige the requests. It wasn’t actually good at doing these issues.” The bot’s responses started altering in 2024, and reached a vital mass late final yr.

In October 2025, she seen that “folks had been placing Halloween apparel on themselves utilizing Grok. After all, a piece of customers realized we are able to additionally ask it to vary what different persons are carrying.” By yr’s finish, “there was an enormous uptick in folks asking Grok to place totally different folks in bikinis or different sorts of suggestive clothes”.

There have been different indications final yr of an elevated willingness to tolerate and even encourage the era of sexually suggestive materials with Grok.

In August, xAI included a “spicy mode” setting within the cell model of Grok’s text-to-video era device, main the Verge to characterize it as “a service designed particularly to make suggestive movies”.

Nwachukwu’s knowledge is simply the newest indication of how the platform underneath Musk has develop into a magnet for types of content material that different platforms work to exclude, together with hate speech, gore content material and copyrighted materials.

On Friday, Grok issued a weird public apology over the incident on X, claiming that “xAI is implementing stronger safeguards to forestall this”. On Tuesday, X Security posted a promise to ban customers who shared little one sexual abuse materials (CSAM). Musk himself mentioned: “Anybody utilizing or prompting Grok to make unlawful content material will endure the identical penalties as in the event that they add unlawful content material.”

Nwachukwu mentioned, nonetheless, that posts like these she has already collected are nonetheless showing on the platform. Musk is giving “the center finger to everybody who has requested for the platform to be moderated”, she mentioned. The billionaire slashed Twitter’s belief and security groups when he took over in 2022.

She added that different AI chatbots do not need the identical points.

“Different generative AI platforms – ChatGPT or Gemini – they've safeguards,” she mentioned. “For those who ask them to generate one thing that appears like an individual, it will by no means be an outline of that particular person. They don’t generate depictions of actual human beings.”

The revelations in regards to the nonconsensual imagery on X has already drawn the eye of regulators within the UK, Europe, India, and Australia.

Nwachukwu, who's from Nigeria, pointed to a selected hurt being finished within the posts to “ladies from conservative societies”.

“There’s a whole lot of concentrating on of individuals from conservative beliefs, conservative societies: west Africa, south Asia. This represents a unique type of hurt for them,” she mentioned.