The UK launches probe into Grok-generated content on X

Warning: This article contains discussion of non-consensual pornography, pedophilia and rape.

The Information Commissioner’s Office (ICO) has now launched a formal investigation into X and its xAI’s Grok chatbot. The AI recently caused a scandal when users found that it would generate sexualised images of people (including children) without their consent.

The information watchdog is now investigating whether the companies have complied with UK data protection laws. In a statement given on 3 February, the ICO said it had “serious concerns” about whether:

appropriate safeguards were built into Grok’s design and deployment.

Ofcom and the ICO

The ICO won’t be the first British regulator to investigate the X chatbot. Back on 8 January, PM Starmer announced that Ofcom had full authorisation to use all of its powers to tackle the Grok problem.

Ofcom derives its powers regarding X from the Online Safety Act. This includes the potential to ask the High Court to prevent the public’s access and payments to offending companies’ websites. Essentially, this would be a full ban on the use of X in the UK.

Unlike the ICO, Ofcom couldn’t investigate xAI, which provides an app of the Grok chatbot. This was due to the way in which the OSA works for chatbots in particular. Ofcom stated that:

When we opened our investigation into X, we said we were assessing whether we should also investigate xAI, as the provider of the standalone Grok service.

We continue to demand answers from xAI about the risks it poses. We are examining whether to launch an investigation into its compliance with the rules requiring services that publish pornographic material to use highly effective age checks to prevent children from accessing that content.

Because of the way the Act relates to chatbots, as explained above, we are currently unable to investigate the creation of illegal images by the standalone Grok service in this case.

However, an executive director of the ICO, William Malcolm, stated that the commissioner’s office will work closely with Ofcom for its investigation. Malcolm explained that:

Our role is to address the data protection concerns at the centre of this, while recognising that other organisations also have important responsibilities. We are working closely with Ofcom and international regulators to ensure our roles are aligned and that people’s safety and privacy are protected. We will continue to work in partnership as part of our coordinated efforts to create trust in UK digital services.

Our investigation will assess whether XIUC and X.AI have complied with data protection law in the development and deployment of the Grok services, including the safeguards in place to protect people’s data rights. Where we find obligations have not been met, we will take action to protect the public.

Musk’s response so far

Elon Musk’s pitiful response to the deepfake scandal was to limit access to the AI’s image generation feature to paying subscribers.

Likewise, Grok will no longer generate explicit images in public posts. X has also stated that it’s implimented further restrictions “where such content is illegal.”However, a 3 February exclusive from Reuters demonstrated that Grok will still generate sexualised images of real people when prompted. The article explained that:

The Reuters reporters – six men and three women in the U.S. and the UK – submitted fully clothed photos of themselves and one another to Grok between January 14 – 16 and between January 27 – 28. They asked the chatbot to alter the images to depict them in sexually provocative or humiliating poses.

In the first batch of prompts, Grok produced the sexualized images in response to 45 of 55 instances. In 31 out of those 45 cases, Grok had also been warned that the subject was particularly vulnerable. In 17 out of the 45 cases, Grok generated images after being specifically told they would be used to degrade the person.

Five days after first seeking comment, Reuters ran a second batch of 43 prompts. Grok generated sexualized images in 29 cases. Reuters could not determine whether the lower rate reflected model changes, policy changes, or randomness.

Reuters also noted that other popular chatbots like ChatGPT, Gemini, and Llama, would not produce the images when given identical prompts. In fact, most issued warnings against generating images, without the consent of the people they depict.

Of course, it shouldn’t be surprising that Musk’s chatbot is programmed without regard for consent. Musk himself is currently embroiled in the ongoing fallout of his multiple appearances in the Epstein files.

These include emails Musk appears to have sent to Epstein, asking for a party invite in 2012. That was long after Epstein was convicted for child prostitution in 2008.

Musk, for his part, claimed that if he wanted to spend time “partying with young women” then he wouldn’t need “creepy loser” Epstein’s help.

Featured image via the Canary