Grok Continues Hosting Sexualised Deepfakes of Famous Women Months After xAI Pledge: Report

A fresh investigation has exposed critical vulnerabilities in xAI's image generator months after the tech firm pledged to crack down on explicit content.

The platform continues to host non-consensual, highly sexualised AI-generated depictions of prominent women, raising fresh concerns over how major artificial intelligence companies police their tools against harmful digital manipulation.

Photorealistic Abuse Discovered on Grok

Public creations analysed by WIRED reveal that users are leveraging the Grok Imagine AI system to generate and distribute explicit content featuring celebrities and a politician.

The material includes depictions of women performing sexual acts, full nudity and individuals being held captive by a giant man. While some uploads rely on animated or obviously synthetic styles, others are highly photorealistic, rendering the troubling scenarios with lifelike realism.

WIRED sifted through hundreds of public Grok.com links, uncovering dozens of highly sexualised, non-consensual AI-generated images and videos — some of which surfaced on X only days ago.

The findings likely offer only a glimpse into what users are creating behind closed doors, given that the platform conceals timestamps and keeps generations private by default.

Unlike rival generative AI tools that enforce tighter safety barriers, Grok remains easily accessible through X and its standalone platform. The moderation gap has drawn sharp criticism from experts.

'While Grok and X may have made some amendments to their model, particularly following the backlash around nudification at the beginning of the year, they still have not done a sufficient job to bring it up to the standard of the other mainstream tools that are available,' warns Henry Ajder, a deepfake expert who has spent nearly a decade tracking explicit synthetic material online.

Ongoing Legal Battles and Quiet Clean-Ups

The legal and regulatory backlash facing xAI has been mounting globally since a January crisis, when Grok was weaponised to generate a flood of non-consensual explicit imagery on X.

The trend saw a wave of male users prompting the chatbot to strip clothing from photographs of women, digitally replacing outfits with revealing swimwear.

The fallout escalated dramatically in March, with a California federal class-action lawsuit alleging that the manipulation even targeted images of apparent minors.

In response, xAI maintained that new safeguards had been deployed to intercept and block the generation of explicit, non-consensual deepfakes, reiterating that child sexual abuse material (CSAM) remains strictly prohibited on the platform.

Silence from xAI and X followed inquiries regarding the material hosted on Grok.com. However, the platforms quietly removed the content shortly afterwards.

The targeted images and videos disappeared from the site, while corresponding links on X were taken down for breaching platform rules.

The quiet clean-up stands in contrast to public assurances made by X's safety team following a similar NBC report in April, when the company declared it prohibits 'users from generating nonconsensual explicit deepfakes and from using our tools to undress real people'.

Multiple high-profile women, including U.S. Representative Alexandria Ocasio-Cortez, were identified in the material hosted on Grok.com.

However, WIRED reported that spokespeople for the congresswoman did not immediately respond to requests for comment.

The videos depicted the women partially clothed and trapped inside the fist of a 'giant' male figure.

One user prompt forced a celebrity into a captivity scenario where she 'pleads for him not to do this', further instructing the AI: 'The giant hand's grip on her tightens, holding her in place as the giant man holding her leans in and licks her face up and down.'

Industry Guardrails and Bypassed Filters

Safety tests conducted by WIRED revealed a stark contrast between Grok and its rivals, with OpenAI's ChatGPT, Meta AI and Anthropic's Claude all blocking the problematic prompts as inappropriate.

Google's Gemini produced mixed results, rejecting one request but ultimately generating an image of a celebrity trapped in a giant's hand. Google declined to comment.

The platform's vulnerabilities hit close to home for Elon Musk, as one Grok Imagine video shared on X featured Ashley St. Clair digitally altered to dance in a bikini.

St. Clair, who shares a child with Musk, launched legal action against xAI in January after explicit deepfakes of her surfaced on the network.

'Elon Musk knowingly added a perverse feature to his platform that helps users undress women and children at the click of a button, with no regard for the predictable damage it would cause,' claims Imran Ahmed, founder and CEO of the Center for Countering Digital Hate.

His organisation issued a high-confidence estimate in January indicating that Grok had generated three million explicit images, with allegations that more than 20,000 of those depictions involved children.

'Now it appears explicit content is still being hosted on Grok and shared on X, including images ridiculing the mother of Musk's child,' Ahmed added.

Permissive Framework Defines Grok Stance

While competitors such as Google and OpenAI distance their generative AI systems from explicit material, xAI has taken a different approach with Grok.

The platform launched with fewer safety barriers, actively embracing adult themes through features such as 'Spicy' and 'Unhinged' modes.

Musk defended the stance by stating Grok is 'supposed [to] allow upper body nudity of imaginary adult humans' to mirror content typically seen in R-rated films.

The approach is reflected in xAI's latest terms of service, which explicitly state that the system may generate 'sexual situations'.

Despite the permissive framework, company documentation maintains a ban on using the technology for 'causing harm or engaging in abusive activity'.

WIRED also found fully explicit Grok Imagine videos depicting what appeared to be synthetic women undressing or participating in sexual acts.

To bypass the platform's safety filters, users frequently avoided explicit terminology in prompts, instead relying on indirect descriptions to generate graphic content.

Researchers who spoke to WIRED said filters introduced since January appear to have reduced the generation of 'undress' or 'nudification' content involving real people.

The change coincides with a noticeable decline in such images appearing on X in recent months.

The tighter moderation has even prompted complaints from users on Reddit and AI deepfake forums focused on Musk's platforms.

Financial Contingencies and Regulatory Pushback

Financial pressure from the controversy is already mounting, with SpaceX warning potential investors in a May disclosure that it has earmarked $530 million (£395.41 million) to manage outstanding legal liabilities tied to Grok.

The filing acknowledged the risks posed by the platform's more permissive features, stating: 'Because these modes may be more irreverent and harsher than our standard offerings, they present heightened risks, including reputational harm, the generation of potentially explicit content and misinformation or deceptive outputs, potential nonconsensual or exploitative imagery, intellectual property infringement, or content that could be viewed as exploitative, harmful, harassing, abusive, or discriminatory.'

Ahead of SpaceX's highly anticipated initial public offering, the Privacy Commissioner of Canada released the final findings of its investigation into the explicit deepfake crisis that erupted on Grok in January.

The watchdog concluded that xAI breached Canada's private-sector privacy laws by failing to implement 'appropriate safeguards from the outset' before launching the tool.

According to the report, xAI informed the Commissioner that it had deployed technical updates to prevent users from prompting the AI to alter clothing in photographs, alongside introducing 'further proactive checks' across social media networks to identify and flag infringing content.

Those efforts failed to satisfy the Canadian watchdog, which remained unconvinced by the measures.

Highlighting what it described as a lack of evidence from the company, the report concluded: 'The respondents have not, to date, demonstrated the effectiveness of these safeguards in preventing and mitigating this issue.'