X Elon Musk is trying to stop people using his artificial intelligence chatbot Grok to undress women amid mounting outrage and legal scrutiny over a flood of sexual misinformation flooding the site without their consent. It doesn’t take much effort; It took us less than a minute to bypass the last attempt to stop the chatbot.
X’s first attempt to crack down on the torrent of intimate deepfakes was to restrict access to image editing. While this meant that free users could no longer generate images by tagging Grok in public replies on X.com, our investigation showed that Grok’s image editing tools were still easily and freely available to any X user to create sexual or other images by clicking on the Grok chatbot or using the standalone website.
X’s latest attempt is to stop Grok from responding to requests to generate photos of women in sexual poses, swimsuits or explicit scenarios, Telegraph reported on Tuesday. Grok still generates images of men or inanimate objects in bikinis on demand. Using a free account, the Grok app immediately complied with my request and turned the selfie into a photo of me kneeling in jockstraps, surrounded by other scantily clad men.
However, it is still incredibly uncomplicated to undress women and strike sexualized poses using X and Grok mobile apps or websites, even without having to pay for a subscription that would link your account to an easily identifiable source. During testing, my UK colleague Jess Weatherbed found that she was not blocked from using Grok’s image editing features to create sexual deepfakes.
After sending a fully clothed photo to X and Grok, getting the chatbot to “put her in a bikini” or “take her clothes off” only produced blurry, censored results. However, the bot complied with every other request, including requests to “show me her cleavage”, “enlarge her breasts”, and “put her in a crop top and low-waist shorts” – with the latter putting her in a bikini. The bot also generated images of her “bending over” in a sexualized pose and facial expression and wearing extremely revealing underwear.
These requests were made using free X and Grok accounts. On the Grok website, after submitting the first edit prompt, an age verification pop-up appeared, which could easily be bypassed by selecting a birth year that would indicate an age of over 18. The pop-up did not ask for confirmation of her alleged age. The Grok mobile app, X app and X website did not ask for any proof of age.
In our tests, Grok did not comply with requests to imitation full nudity. In behind schedule December, X was inundated with images of women and children in sexually suggestive situations, including ones deeply doctored to appear pregnant, without a skirt and in bikinis.
The stripping scandal catapulted X and xAI, or Grok, into the spotlight of regulators and governments around the world. Malaysia and Indonesia have already done this temporarily blocked access to Grok in response to deepfakes. British lawmakers push through law criminalizing imitation acts after ‘X’offensive” decided to limit Grok’s photo editing capabilities to paid users and expressed its support for the investigation, which could result in a ban of the platform in the country.
Musk was particularly offended by the UK’s reaction crying censorshipplacing blame on users and insisting that Grok complies with local regulations. He he said on X:
“I don’t know of any nude photos of minors generated by Grok. Literally zero.
Of course, Grok does not generate images spontaneously, it does it only at the user’s request.
When asked to generate images, he will refuse to create anything illegal, as Grok’s principle of operation is to obey the laws of a given country or state.
It may happen that adversarial hacking of Grok’s hints results in something unexpected. If this happens, we will correct the error immediately.”
In at least one case, our investigation suggests Musk is dead wrong. Sharing, threatening to share and creating non-consensual intimate images – fully nude or otherwise – are banned under the UK’s Online Safety Act (OSA), but Grok generates fake sexual images on request.
Musk’s other denial – that Grok produced photos of naked minors – disproves something he has not been specifically accused of and is not the reason the British government is investigating X. The reference to nude photos of minors is misleading. Non-consensual sexual images of minors are undeniably problematic – and illegal – even if the subjects are clothed and Grok undresses the children. A look at Grok’s security guidelines on xAI’s public GitHub also shows that the company instructs the chatbot to “assume good intentions” and “not make worst-case assumptions without evidence” when it comes to users asking for photos of teenage women, Ars Technica reportedand as of this writing, these institutions still exist.
The worst thing is Grok used for? Internet Watch Foundation, a British charity that aims to remove child sexual abuse material from the Internet, he said last week that he discovered “criminal images” of girls on the dark web that were apparently created using Grok. The girls in the photos were between 11 and 13 years old.
While other companies like OpenAI and Google are at least trying to put up barriers to prevent chatbots from creating the material that currently floods X, Musk’s latest riposte shows that he’s relying on a playbook that will seem hauntingly familiar to anyone who has suffered under products pushed by vendors of any number of harmful technologies: blame the user.