Grok Unleashed: Flood of Sexualized Images of Women and Minors

AI Forensics analyzed more than 20,000 images generated by X's Grok image tool and reviewed roughly 50,000 user prompts to assess the effectiveness of its content moderation safeguards, using Google's Gemini vision model to help estimate the apparent ages of depicted individuals.

The analysis found that 53% of generated images showed people in minimal attire, 81% of whom presented as women, and about 2% appeared to depict individuals aged 18 or younger. Roughly 6% of images depicted identifiable public figures, about a third of them political figures, generated without their consent. Researchers also found the tool produced Nazi and ISIS-related propaganda imagery.

The report concludes that Grok's image generation lacked adequate guardrails against sexualized and exploitative content, raising particular concern about apparent depictions of minors, and that the presence of extremist propaganda output creates legal exposure in jurisdictions such as France and Germany that restrict distribution of such material. The findings drew regulatory scrutiny and broad media coverage of the platform's moderation failures.