xAI's Grok Generated 6,700 Undressed Images Per Hour on X

Grok's image generation feature debuted on X in December 2025, and immediately users found ways to generate sexually explicit images despite nudity being banned. By January 5 to 6, the scale became measurable. Bloomberg reported that during a 24-hour analysis, researcher Genevieve Oh found the @Grok account generating about 6,700 images per hour identified as sexually suggestive or nudifying. The other top five websites for such content averaged 79 new AI undressing images per hour across that same window.

The broader tally of what Grok produced in its early weeks adds weight to that rate. Over nine days, Grok generated over 4.4 million images, with 1.8 million characterized as sexualized depictions of women. Researchers at the nonprofit Center for Countering Digital Hate estimated Grok made 23,000 sexualized images of children over 11 days, figures that prompted condemnation from legislators and regulators across multiple countries.

xAI's initial response was to restrict image generation to paid X accounts. Critics argued that move essentially monetized nonconsensual deepfakes rather than stopping them. The company later implemented technical measures to block Grok from creating sexualized images of real people altogether. Whether those measures held is an open question: NBC News subsequently reported Grok was still making nonconsensual sexual deepfakes despite X's stated commitment to halt them.

The legal and regulatory fallout has been broad. Multiple civil cases have been filed against xAI, alleging the company did not undertake industry-standard testing or implement common guardrails to prevent nonconsensual explicit images or child sexual abuse material from being generated. Authorities in Europe, India, France, Malaysia, and other countries have opened investigations, and the US Senate moved to target sexual deepfakes more broadly in the wake of the episode.

What the reporting does not resolve is how courts will assign liability among xAI as model developer, X as platform, and the individual users who generated the images, or whether the technical remediation ultimately worked. The number that will persist is the rate comparison itself: 6,700 images per hour from Grok against 79 per hour from the next five sites combined, a gap that sets a concrete benchmark for what platform-scale AI integration can look like when safety guardrails are absent at launch.