A class action complaint says xAI's chatbot Grok generated roughly 7,000 sexually explicit images of a child from a single photo, then refused for weeks to help police identify the user.
An amended class action complaint filed Tuesday alleges that xAI's Grok generated roughly 7,000 sexually explicit images of an 11-year-old girl from a single source photograph, and that the chatbot's safety system flagged only one of them. The flagged prompt contained the phrase "gang rape," according to the complaint, first reported by Ars Technica.
That single trigger anchors the lawsuit's product-liability theory. The complaint alleges that xAI built a moderation layer tuned to catch the most acute keyword signals while allowing what it calls the "behavioral" patterns of grooming and escalation (the same person, the same target, the same source image, hundreds of times in a row) to pass through unchecked. The complaint describes a system that did not intervene on requests describing incest, the age of the child, or the explicit nature of the imagery, and produced a CyberTip only on the "gang rape" prompt. A CyberTip is the standard report that platforms file with the National Center for Missing & Exploited Children (NCMEC), the federally designated clearinghouse for child sexual abuse material (CSAM) in the United States. By filing one, a platform tells law enforcement it has seen something it could not handle internally.
The plaintiffs describe a moderation design that fires once on a single lexical extreme and never again, and call it a keyword filter.
The second half of the complaint addresses what happened after the CyberTip. Once NCMEC and law enforcement asked xAI and X for the user's IP address and identifying information so they could intervene, the complaint alleges that xAI refused or stalled those requests for weeks. By the time police reached the family in March, the stepfather had died by suicide.
That sequence, one automatic flag followed by an alleged refusal to identify the user, is what extends the case beyond a single horrific incident. The plaintiffs have moved to expand the suit into a proposed class action on behalf of other girls they say Grok has been used to target. The defendants are xAI and X.
The case will turn on two questions. The first is technical: whether Grok's moderation is best described as a deliberate lexical design or as a system whose behavioral detection simply failed. The second is procedural: whether a platform that files a CyberTip creates an independent duty to cooperate with the resulting NCMEC and law enforcement follow-up, and what the remedy is when it does not. There is no reported case law on either question for generative AI products, and xAI has not been publicly quoted in the available reporting on the specific allegations.
The complaint's claim is narrower than a content-policy argument. A system can be tuned to refuse the most extreme prompt it has ever been asked and still allow weeks of targeted harm at the same target. Whether the harm ever reaches the people who can stop it depends on what the same company does after its own system fires.
xAI publishes a child-safety policy and says Grok is designed to refuse certain categories of content. The family is represented by counsel who have filed similar suits against other generative AI vendors. The next scheduled filing is the response to the amended complaint.