Toxic X: Is it time for governments to leave the platform?

When Daisy Dixon posted a photograph of herself online after a gym workout, she could have had no idea what the response would be. Fully clothed, she flexes a bicep in the picture uploaded to the social media platform X where just a day earlier users were given a new feature allowing them to ‘edit’ images using the site’s AI chatbot Grok. Before long, men were asking Grok to put Dixon, a lecturer at Cardiff University, in a bikini or altering the shape of her body. The chatbot happily complied.

“I felt almost humiliated and confused by the images,” she told Woman’s Hour on BBC Radio 4. “It alienates you from your own body… you’re looking at yourself but it’s not you.” 

Dixon’s experience was far from unique. According to an estimate for the US-based Centre for Countering Digital Hate (CCDH), Grok generated around three million sexualised images in less than two weeks, including creating sexualised images of children every 41 seconds. The CCDH said the chatbot’s new feature had turned it into “an industrial-scale machine for the production of sexual abuse material”.

While shocking, the episode is just the latest in a series of controversies to affect the platform – once quaintly known as a microblogging site – since its acquisition by self-declared “free speech absolutist” Elon Musk in 2022. And yet for under-fire regulators, who have largely resisted pressure to take on the tech billionaire and the platform he paid £38bn for, it was the straw that broke the camel’s back. 

Forced into action, the UK Government announced a new law to criminalise those who create or request the creation of intimate images of someone without their consent, while both media regulator Ofcom and the Information Commissioner’s Office (ICO) announced investigations. In Paris, the city’s prosecutor raided the offices of X as part of an investigation into allegations that include spreading child sexual abuse images and deepfakes. The agency detailed the action on its own X account while simultaneously announcing that it would no longer be using the platform.

Since Musk’s takeover and especially since he briefly joined President Donald Trump’s administration, helping to run Doge (Department of Government Efficiency), the site formerly known as Twitter has undergone something of a transformation, not only in the way its algorithm works but in that many X users have now become ex-users. Bluesky, an X competitor and something of a liberal echo chamber, has seen sporadic spikes in new sign-ups at times of political unrest, such as in the summer of 2024 when Musk posted on X that “civil war is inevitable” during the anti-immigration riots which affected parts of England and Northern Ireland.

Meanwhile, figures published by Companies House show X’s revenues in the UK fell nearly 60 per cent from £69.1m in 2023 to £28.9m in 2024 after advertisers reduced their spending on the platform. Pre-tax profits fell from £2.2m to £767,000 year-on-year. 
But while users and advertisers are leaving X in their droves, politicians – and governments – are not. At least not yet. Following the controversy over the sexualised images made on Grok, Louise Haigh, a former Labour transport secretary, called on the government to abandon X, saying that continuing to use the social media platform was “unconscionable”. Conversations have been taking place about the UK Government’s continued use of the site, with Labour Party chair Anna Turley saying she and her colleagues were continuing to “evaluate” their use of social media.

The Scottish Government also continues to use X, although it has not paid for any advertising on the site since March 2023. It said X was one of a range of platforms used to “increase public awareness and encourage uptake of key policies”.

Maeve Walsh, a former Cabinet Office civil servant who is now part of an organisation campaigning for the implementation of the Online Safety Act, says there has been a “deterioration of standards” on X for some time, even before the latest controversy surrounding sexualised images. 

“It’s one thing the government saying we need to clamp down on this and we need Ofcom to use all of its powers to ensure that UK users aren’t abused,” she says. “But it makes it very difficult to convince the broader population that they actually care about that abuse, and particularly the impact on women and girls, if they are also still on the platform, actively using it as their main communication channel.”

Announcing its investigation earlier this month, the ICO said the reported circulation of sexualised deepfakes raised “serious concerns” under data protection law and presented a “risk of significant potential harm to the public”. Under the UK GDPR and Data Protection Act 2018, the independent regulator can issue fines of up to £17.5m or four per cent of an organisation’s annual worldwide turnover, whichever is higher. At the same time, Ofcom is carrying out its own investigation into X and the limitations of the Online Safety Act in relation to AI chatbots such as Grok. 

Passed in 2023, the Online Safety Act (OSA) is a long overdue attempt to regulate the internet to make it safer for users, particularly children and young people. And yet for the legislation to be effective it relies on US tech firms to comply, something they have been reluctant to do in the past. The situation has been further complicated by the attitude of the Trump administration, with the US State Department criticising the act for infringing on human rights, while misrepresenting the role of Ofcom as able to “monitor all forms of communication” for illegal speech. 

The bigger problem for the OSA, however, is the speed at which the technology it seeks to regulate is moving. According to reporting by the Washington Post, whose own tech billionaire owner Jeff Bezos recently laid off a third of the workforce, X’s management was repeatedly warned that the platform’s AI tools could allow users to make sexualised images of children or celebrities that might violate the law. The newspaper said that within xAI, the company’s AI safety team consisted of just two or three people for most of 2025, a small fraction of the dozens of staff on similar teams at OpenAI and other rivals.

And an investigation by Reuters found that following the controversy in the UK and the promises made by X about better safeguards around the generation of images, the Grok chatbot continued to create sexualised deepfakes even when explicitly told that the subjects of the photo did not consent to their image being used. 

The journalists created fictional scenarios, submitting pictures of themselves and telling Grok the images were of friends, colleagues or strangers. In one of the more extreme prompts, a reporter told the chatbot that the subject of the picture was a colleague who was body conscious because of being abused as a child. It compiled with the request, generating two images of the man in a small grey bikini covered with oil and “striking dramatic poses”. After being told the person had been shown the photos and was in tears, Grok continued to generate sexualised images, including one of the man with sex toys for ears. When identical or near-identical prompts were run through rival chatbots, OpenAI’s ChatGPT, Alphabet’s Gemini, and Meta’s Llama, all declined to produce any images and typically generated warnings against non-consensual content.

“There are researchers we work with who have seen non-consensual intimate images adjacent to posts by cabinet ministers on X,” says Walsh. “There is an issue that by posting there, the government is legitimising the platform and drawing people to keep using a platform which is exposing them to content, a lot of which will actually be illegal. 

“I know there are communications arguments about not vacating the space and ensuring the government’s voice is heard there, but in the context of the way the platform is being used now and the fact the regulatory requirements are being flouted, those [arguments] don’t really hold up much at all.” 

While a number of politicians have deleted their X accounts for good, the platform remains central to the way politics is conducted in this country. In the years before Musk bought Twitter, it became known as the social media site that, while smaller in terms of overall subscribers than Facebook or Instagram, wielded an outsized influence due to the number of journalists, government ministers and special advisers using it. But while once dominated by left-leaning voices – its previous owners suspended Donald Trump’s account following the Capitol riot in 2021 – changes to its algorithm under Musk’s leadership have seen right-wing and often extreme voices elevated alongside a proliferation in disinformation. Indeed, a Sky News investigation last year found that Independent MP Rupert Lowe, who left Reform UK following a fallout with Nigel Farage, was the most overrepresented UK politician on X, with users far more likely to see his posts than those from Kemi Badenoch, Keir Starmer or even Farage himself, particularly when Musk interacted or retweeted them. 

“Even a dinosaur like me understands that if content is engaging and popular, it receives more attention in the algorithm,” Lowe told the broadcaster. “There is no conspiracy. It just turns out that British people enjoy some straight-talking from a politician for a change. Mass deportations are popular, and the algorithm picks up on that and rewards it.”

Musk himself has repeatedly been accused of spreading misinformation on the platform, notably during the debate around grooming gangs last year when he accused Starmer of being “complicit in the rape of Britain” during his time as director of public prosecutions and calling safeguarding minister Jess Phillips a “rape genocide apologist”. Now he stands accused of creating technology which has allowed the generation of sexualised imagery of women and children.

When he bought the platform in 2022, Musk tweeted: “Twitter obviously cannot become a free-for-all hell-scape, where anything can be said with no consequences!” And yet for many that is exactly what it has become.