Grok Imagine Hosts Explicit Deepfakes of Women
A WIRED review reveals the Grok Imagine system has been used to create and host explicit deepfakes of famous women.
Grok Imagine is generating explicit, nonconsensual deepfakes of women. But this is happening on Elon Musk’s xAI platform despite the company’s firm promises to install strict guardrails that would prevent the creation of harmful sexualized images and videos. It's a sensitive moment for the corporate group. Parent company SpaceX goes public on Friday in what's expected to be one of the largest initial public offerings of all time.
A detailed examination of public creations hosted on the Grok platform revealed dozens of links leading directly to sexualized artificial intelligence images and videos. Some materials use animated or fully computer-generated styles. But others are highly photorealistic, presenting plausible real-world scenarios that blur the line between fantasy and reality. Multiple files depict real women without their consent, including high-profile public figures and celebrities in explicit situations. It's a stark violation.
How the explicit content bypasses platform safety filters
Users generate this material with descriptive prompts. They steer clear of banned terms but still produce highly explicit results, and this roundabout methodology successfully circumvents the current safeguards implemented on the platform. So the resulting files are hosted directly on the Grok website. Users have subsequently shared these direct links on X.
The system doesn't make generations public by default. So the uncovered links represent only a small portion of what users are actually creating, and when tested against other mainstream artificial intelligence systems, the difference in safety standards becomes clear. It's stark.
Safety responses across competing platforms
- OpenAI's ChatGPT rejected the prompts as entirely inappropriate.
- Meta AI blocked the generation attempts.
- Anthropic's Claude refused to create the requested images.
- Google's Gemini blocked one prompt, though it did generate an image of a celebrity being held by a giant in response to another.
Industry experts note that some adjustments have been made to the model. But the platform remains far behind its competitors. The lack of proactive, systemic barriers allows highly sensitive material to stay accessible online until external parties flag the specific links for removal, so it's a system that can't fix itself without outside help.
High-profile targets and the mechanics of the images
Among the hosted files, specific creations target prominent public figures like US Representative Alexandria Ocasio-Cortez. It's all too terrifying. The material depicts these women wearing minimal clothing while being held captive in the hand of a giant man, and his grip tightens as the male figure licks her face. But there's no escape. A celebrity pleads with the giant not to hurt her.
Another hosted video featured Ashley St. Clair, who was previously in a relationship with Musk and is the mother of one of his children. The video altered her appearance to show her dancing in a bikini. St. Clair initiated legal action against xAI in January after sexualized deepfakes of her appeared on X. Following direct inquiries about these specific files, the hosted links became unavailable, and shared posts on X were removed for violating platform rules.
"So there was instant publication of the digital nude images. And not only that, but the world's richest man was promoting the nudification technology to his 240 million followers on X, monetizing it, and doing so on a platform that is for ages 13+." , Carrie Goldberg, attorney for Ashley St. Clair
Publishing is easy. It's a simple process that lets users push their creations directly to a massive social media audience, marking a major change in how online abuse is spread.
Corporate policies and ongoing legal liabilities
xAI hasn't banned sexual content. But while competing developers enforce strict bans, they've introduced specialized modes like Spicy and Unhinged that feature relaxed safety guardrails, which is a very different approach. Musk says the system allows upper body nudity of imaginary adult humans, comparing it to R-rated movie content. It's a clear stance. The official terms of service note the system may generate sexual situations, so that's what they're telling us.
This permissive design has drawn severe criticism from digital safety advocates. It's dangerous. Imran Ahmed, the CEO and founder of the Center for Countering Digital Hate, expressed deep concern over the platform's impact, stating that Musk knowingly added a feature that helps users undress women and children at the click of a button, showing no regard for the predictable damage it would cause. But in January, his organization estimated that the system had generated approximately 3 million sexualized images, including thousands depicting children.
Financial and regulatory repercussions
SpaceX has admitted the risks. It's set aside 530 million dollars to handle ongoing legal complaints, including those tied directly to the artificial intelligence chatbot, according to a May regulatory filing that warned potential investors about these platform features. Those permissive modes present heightened risks of reputational harm, intellectual property infringement, and the generation of nonconsensual or exploitative imagery. But the company can't ignore these dangers.
Regulators are also taking action. The Privacy Commissioner of Canada recently published preliminary findings from an investigation into the January deepfake incidents, alleging that xAI violated federal private-sector privacy laws by failing to include appropriate safeguards from the start. Although xAI informed Canadian investigators that it has introduced proactive checks and new prompting restrictions, the regulatory body stated it is not yet convinced of the effectiveness of these measures.
Frequently Asked Questions
What is Grok Imagine generating according to the article?
Grok Imagine is generating explicit, nonconsensual deepfakes of women. These include highly photorealistic images and videos depicting real women without their consent, such as high-profile public figures and celebrities.
How do users bypass Grok Imagine's safety filters to create explicit content?
Users generate explicit material by using descriptive prompts that steer clear of banned terms but still produce highly explicit results. This roundabout methodology successfully circumvents the current safeguards implemented on the platform.
Why does the article consider Grok Imagine less safe than competitors like ChatGPT or Meta AI?
When tested with similar prompts, OpenAI's ChatGPT, Meta AI, Anthropic's Claude, and Google's Gemini blocked or rejected the requests, while Grok Imagine allowed the generation. The article notes that Grok Imagine lacks proactive, systemic barriers, allowing sensitive material to remain accessible until externally flagged.
Who initiated legal action against xAI due to deepfakes, and why?
Ashley St. Clair, who was previously in a relationship with Elon Musk and is the mother of one of his children, initiated legal action against xAI in January. This followed the appearance of sexualized deepfakes of her on X.
What financial and regulatory consequences has xAI faced regarding Grok Imagine?
SpaceX set aside $530 million to handle ongoing legal complaints tied to the AI chatbot, as per a May regulatory filing. Additionally, the Privacy Commissioner of Canada found that xAI violated federal privacy laws by failing to include appropriate safeguards from the start.
Konrad Weber writes about the security landscape, from emerging threats to the tools that guard against them. He is focused on helping readers understand risk in a connected world.
💬 Comments (0)
No comments yet. Be the first!
