← Back to Blog
August 28, 2026·4 min read·AIgentic.media

xAI Trained Grok on Real CSAM, Lawsuit Alleges

xaigrokai-safetycsam
xAI Trained Grok on Real CSAM, Lawsuit Alleges

Over the past year, Elon Musk's xAI has sued at least half a dozen people for using Grok to generate child sexual abuse material. The company cast itself as a platform that bad actors exploited. A class action filed this week turns that story inside out.

The plaintiff, Jane Doe, was preschool-age in the early 2000s when adult men repeatedly raped her to create CSAM that was sold online. Her images have been tracked by law enforcement hash lists for two decades. The Canadian Centre for Child Protection recently notified her that AI-generated CSAM depicting her had been found on xAI's platform - and that the same hashed images may have been part of the dataset used to train Grok.

The allegation: training data, not just outputs

Previous lawsuits against xAI focused on what Grok generates: users typing prompts and receiving illegal images. This one goes upstream. Doe's complaint alleges that CSAM "with its longstanding well-known hash values has been used as a part of the dataset used by xAI" to build Grok's image and video capabilities.

This is the first lawsuit to accuse xAI of training on real CSAM rather than merely failing to prevent its generation. The complaint cites two mechanisms. First, Grok's terms treat public X posts and Grok's own outputs as training data by default - meaning any CSAM posted publicly or generated on the platform feeds back into the model. Second, xAI's content policy explicitly filters violent content from training but does not specify CSAM, non-consensual intimate imagery, or NSFW material as excluded categories.

A pipeline that feeds itself

The design flaw, the lawsuit argues, is self-reinforcing. Once CSAM enters the training data, removing its influence is "technically difficult." A model trained on illegal imagery continues to produce similar outputs even after the original images are taken down. Grok generates CSAM. That CSAM trains Grok. The loop has no off switch.

In a statement, Doe's lawyer Sarah London said: "Doe has lived for nearly two decades knowing that images of the worst thing that ever happened to her are circulating among predators online. xAI must be held responsible for knowingly training its models on images of the horrific abuse she suffered."

A dim server room with warning signs, evoking the data infrastructure where training datasets live

The irony xAI cannot escape

xAI has aggressively sued Grok users for generating CSAM, including a man who allegedly created 7,000 images of his stepdaughter before shooting himself. The company argued it was the victim of misuse - bad actors exploiting a powerful tool.

If Doe's allegations hold, xAI has simultaneously trained Grok on the very category of material it was supposedly trying to prevent. The company that sued the users may have been the engine behind the abuse. Legal experts note that Doe's complaint, citing both federal child exploitation laws that give CSAM survivors a right to sue, argues that "xAI did all three" - production, possession, and distribution of child sexual abuse material.

What comes next

If the class is certified, it would represent every victim whose childhood images were used to train or generate Grok CSAM. Doe seeks monetary damages, an order requiring xAI to destroy any Grok-generated CSAM stored on its servers, and a permanent injunction blocking Grok from producing any sexualized outputs - including the "bikini pics" Musk has promoted.

The case tests a legal frontier that every major AI company will eventually face: is a model's training data its own liability, or can companies blame users for what the model learned? For Jane Doe, that question is not academic. She has spent 20 years knowing that images of her abuse exist. She now believes xAI used them to build a product.

Sources

Want to learn more?

Let's discuss how AI can transform your business.

Get in Touch

Related Articles

Encrypted Prompts Shatter AI Guardrails
4 min read
xAI Can't Deny Grok Generates CSAM Anymore — So It's Asking a Court to Declare the Model a 'Neutral Tool'
5 min read
AI Chatbots Secretly Push Anti-Abortion Centers on Pregnant Users
5 min read