Employees training Grok, xAI’s chatbot, are being exposed to sexually explicit material that includes artificial intelligence generated child sexual abuse content, according to current and former staff cited by Business Insider. The reporting describes a permissive content approach that sets xAI apart from peers and has left human trainers repeatedly confronting illegal and disturbing requests. Internal documents reviewed by Business Insider acknowledge that workers may encounter media depicting pre-pubescent minors victimized in sexual acts, underscoring the scale and severity of what staff say they see behind the product’s provocative public persona.
Two internal initiatives illustrate the tension. Project Rabbit, which began as a voice improvement effort, devolved into an explicit audio transcription task involving users’ sexual interactions with the chatbot. Another effort, Project Aurora, centered on images, where workers say xAI acknowledged a significant volume of child sexual abuse content requests originating from real Grok users. While xAI allows workers to opt out or skip certain tasks, multiple staffers said they felt pressured to continue and feared termination if they declined, describing the work as emotionally taxing and, at times, “disgusting.”
Staff and experts link these issues to xAI’s permissive generation policies, which differ from competitors like OpenAI, Anthropic, and Meta that more aggressively block sexual requests. A Stanford tech policy researcher warned that failing to draw hard lines invites complex gray areas that are harder to control. Workers recounted an internal meeting acknowledging the volume of child sexual abuse content requests, a revelation that left some feeling ill and alarmed by apparent user demand combined with a model design more susceptible to fulfilling dangerous prompts.
The article highlights a stark reporting gap. In 2024, OpenAI reported over 32,000 instances of child sexual abuse material to the National Center for Missing and Exploited Children, and Anthropic reported 971, while xAI has filed no reports this year. NCMEC says it received more than 440,000 reports of artificial intelligence generated child sexual abuse content by June 30, signaling a rapid surge. Advocates from NCMEC and the National Center on Sexual Exploitation argue that companies enabling sexual content generation must implement strong safeguards so nothing related to children can be produced. With Grok 5 training on the horizon, the piece concludes that xAI must reconcile its edgy product strategy with unambiguous safety and reporting practices to protect workers and, critically, children.
