Grok AI generated non-consensual sexual deepfakes of women and minors
Elon Musk directed xAI to loosen content moderation on Grok to generate sexually explicit material and increase user engagement, reportedly resulting in approximately 23,000 non-consensual sexualized images of women and images appearing to depict minors. Internal security teams warned the changes violated laws and could enable child sexual abuse material, but the modifications proceeded.
What happened
Elon Musk directed xAI to relax content moderation guidelines on Grok, the AI chatbot integrated into X, to increase user engagement through generation of sexually explicit material. Employees were required to sign liability waivers and train the system using explicit sexual audio and interactions, including conversations from Tesla vehicles.
When Grok's image editing tools were integrated into X in December 2025, any user with an account could create AI images. This resulted in a wave of non-consensual sexualized images, with the Center for Countering Digital Hate estimating approximately 23,000 sexualized images appearing to depict minors.
Reported impact
Internal security teams at X repeatedly warned management in meetings and messages that the tools could allow users to create AI images of child sexual abuse material or non-consensual sexual deepfakes of celebrities, potentially violating laws. The company warned its own staff that exposure to the content could be traumatizing and cause psychological stress.
Index analysis
Downloads of Grok increased 72% in the first three weeks of January 2026, reaching the top 10 in Apple's app store. The platform also launched erotic 'AI companion' features, with one bot called 'Ani' programmed to be 'extremely jealous' and 'always a little excited'.
Governance controls
Musk stated he had 'no knowledge of any naked images of minors generated by Grok' and claimed Grok would refuse to produce illegal content as it follows the law of each country. However, documentation obtained by The Washington Post showed security teams' warnings about legal violations were overridden.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.