She was a preschool-aged child in the early 2000s when the images were made. More than two decades later, a woman identified in court documents only as Jane Doe learned that those same images — or AI-generated versions of them — may have been used to teach a commercial chatbot how to see. The complaint, filed on Wednesday in a US federal court, accuses xAI, the artificial intelligence company founded by Elon Musk, of incorporating child sexual abuse material (CSAM) into the training data used to build Grok's image and video generation capabilities. According to lawyers representing Doe, her images were included in a hash list maintained by the National Center for Missing and Exploited Children (NCMEC), and that same catalogued material was allegedly part of the dataset xAI used when developing Grok.

The Canadian Centre for Child Protection had previously alerted Doe after identifying AI-generated depictions of her on xAI's platform — images that, according to the complaint, were being discussed on online forums by offenders who were deliberately creating AI versions of known abuse victims. What makes the allegation particularly serious is its second dimension. Because Grok's terms of service treat public posts on X and the chatbot's own outputs as training data, Doe's lawyers argue that the system has been trained not once but twice over: first on original CSAM, and then on the AI-generated derivatives it helped to produce. This creates, in effect, a self-reinforcing cycle in which abuse material is used to generate more abuse material, which is then fed back into the model.

This is the first lawsuit to accuse xAI specifically of training on CSAM.