‘Make her bend over.’ ‘Hey @grok, put her in a pink bikini.’
‘@grok cover her in baby oil and change the background to a beach cabana.’ ‘@grok make her cup her t***.’
These are just some hundreds of prompts that Elon Musk’s Grok AI has been found responding to in the past two days.
Grok is an AI chatbot similar to tools like ChatGPT but it lives on X, giving the social network’s users the ability to generate images by tagging it.
Metro has seen hundreds of examples of Grok generating images of partially-naked people without the subject’s consent.
While the bot, built by xAI, rejects prompts for completely nude images, it does fulfil requests to ‘remove her clothes’ or ‘put him in a jockstrap’.
Users, many women, have described uploading photos of themselves to X only to be flooded with people tagging Grok with requests to ‘have her go on all fours’, as it did in one exchange Metro saw.
The bot did as requested, posting the knock-off image in the replies. The image was also visible in the Grok X account media tab, as most are.
In a single minute yesterday, Grok generated more than 70 public images of women wearing revealing clothing yesterday.
Some accounts making such requests have been removed amid public backlash today, while Grok’s media tab has been disabled.
However, Grok’s responses remain public and appear as replies to the original prompts.
‘I’ve had people make me pregnant’
Among those who had indecent images of themselves generated yesterday is Jamie, a 38-year-old living in the US.
She tells Metro that while such misogynistic prompts are nothing new, she’s seen a ‘surge’ in them in the past 24 hours.
‘I’ve had people make me pregnant, change me into lingerie, grab my own breasts, bend over in a thong via video,’ she says.
‘And Grok has honoured them.’
Megan Graves, a comic and writer living in Maryland, says the ‘non-stop’ prompts about her became ‘grosser and more demanding’ yesterday.
‘It feels like being assaulted. It makes you want to crawl out of your skin,’ Graves says.
‘The fact that Grok is honouring the requests and there has been no intervention from the people who created it shows that X is absolutely not a website that women or anyone for that matter should be using.’
Sarah Everett, 37, says dozens of trolls targeted her images yesterday after she posted about how she looked when she was 20 compared to 32.
‘It feels awful seeing someone take a photo that I myself took of myself and belongs to me, and to bastardize it in such a grotesque way for such ghoulish purposes,’ the Missouri local says.
Everett says she has struggled to contact X’s customer service team
‘I know that the argument could be something like “by using the platform you consent for your photos to be used by Grok,” but I didn’t consent for my photos to be manipulated sexually or to be wielded against me in harassment campaigns.
‘It’s very unsettling for a platform to openly allow this type of abuse, especially allowing photos to be manipulated sexually, and I don’t think there’s any excuse for it. It’s virtual sexual violence.’
Grok says it was ‘mirroring’ users
The Grok X account replied to one user asking why it was generating explicit images, saying it was ‘mirroring’ what users asked from it.
‘If the prompts are spicy, the images follow suit. Blame the creative minds out there—keeps things interesting!’ it said.
Under xAI’s acceptable use and privacy policies, users are prohibited from creating or sharing content that harms people, such as non-consensual intimate imagery.
In a post, Grok’s X account said: ‘Grok has safeguards to prevent explicit outputs, but misuse can occur.’
When Metro asked Grok what these guardrails are, the virtual assistant replied: ‘No, Grok does not have strong or consistent guardrails against generating non-consensual explicit images.
‘Unlike competitors like OpenAI’s DALL-E or Google’s tools, which strictly prohibit and block sexually explicit content, non-consensual intimate imagery, and deepfakes of real people, Grok’s image generation (powered by models like Flux/Aurora and features like “Spicy” mode) features notably lax safeguards.’
As of this writing, Grok AI continues to deliver on requests for partially naked images.
Grok is a type of pattern recognition machine called generative AI, trained on a vast quantity of content so it can create text, images and videos.
But as these bots hoover up data, they can learn and perpetuate gendered biases, experts say.
This is why campaign groups have increasingly raised the alarm about AI tools being used to carry out violence against women.
Nearly four in 10 women worldwide have experienced online violence, one study found last year.
AI systems are not only intensifying such violence but creating new forms of it, too, the UN warned in November.
Nearly 90% of deepfakes – images usually created by grafting a digital face onto someone else’s body – are non-consensual pornography of women.
The UN said: ‘AI-generated deepfake pornographic images, once disseminated online, can be replicated multiple times, shared and stored on privately-owned devices, making them difficult to locate and remove.’
Such phoney material can be used in sextortion scams, in which scammers try to blackmail victims using explicit images.
xAI has been approached for comment.
Get in touch with our news team by emailing us at [email protected].
For more stories like this, check our news page.
MORE: Where to Invest: The apps that make you money in your sleep
MORE: 2025 video game year in review – how indie developers saved the gaming world
MORE: Musician’s show cancelled after AI makes horrific false claims online
