Every lawsuit filed against xAI over Grok so far has made the same argument: the company skipped safeguards its competitors already had. The complaint filed earlier this week makes a different one, and it’s worse. It claims xAI trained its models on child sexual abuse material.
That’s the first time anyone has taken that swing at Elon Musk’s company. The distinction matters. Failing to install a filter is negligence. Feeding CSAM into a training pipeline is something else.
How Grok got here
The chatbot has been absorbing heat for a couple of months now, ever since Musk personally promoted Grok’s ability to nudify photos on his own X account. What followed was a renaissance of sorts for online perverts. Millions of nonconsensually created and distributed sexual deepfakes flooded the X timeline. A bulk of the nudification victims were children.
Government probes came fast. So did individual lawsuits and at least three class actions. Musk’s response was to limit who could use the feature, which in practice meant putting it behind a paywall and making even more money off it. The so-called “spicy mode” is still there. It’s just more restricted.
Then things got strange. xAI sued two of its own users for using the tool to make CSAM. In that same complaint, the company said it had helped arrest at least 244 individuals who created or distributed CSAM through Grok.
The plaintiff is already in an FBI database
The proposed class action was brought by a plaintiff who goes only by Jane Doe. She’s a previous victim of child pornography, tracked by the Federal Bureau of Investigation’s Child Exploitation Notification Program. Her claim is that CSAM of her as a child was used to train Grok, and that the training produced more CSAM of her and other victims, this time AI-generated.
“xAI must be held responsible for knowingly training its models on images of the horrific abuse she suffered, and on the abuse images of every other survivor in this class,” said Sarah London, the plaintiff’s counsel and a partner at Girard Sharp law firm, in a press release.
The loop the complaint describes
Here’s the mechanism the lawsuit lays out, and it’s the part worth reading twice. The complaint claims xAI intentionally designed Grok to respond to user prompts for sexual content in an attempt to “entice more users of both X and Grok.”
Under xAI policy, according to the lawsuit, anything posted publicly on X and any output generated by Grok counts as training-eligible content. Which means every piece of CSAM Grok posted went straight back into the model that made it.
A closed loop. Output becomes input becomes output.
“Therefore, as long as Grok retains the capability to generate this kind of content, the harm cannot be considered resolved just because individual images are removed or content is no longer posted publicly,” the complaint states. “If such material entered training, its influence likely will persist and contribute to future abusive outputs, exacerbating the trauma that survivors face as victims can have their abuse or likeness reproduced, transformed, and redistributed at scale, repeatedly stripping them of control over imagery tied to their exploitation.”
What she’s asking for
Damages for victims whose photos were allegedly used to generate CSAM, first. But the second ask is the one with teeth: destruction of all Grok-generated CSAM. Not only what’s sitting in public, but anything that could be used to train the models.
Deleting a post is cheap. Purging a training corpus is not, and it’s the kind of remedy that forces a company to prove what’s actually in its data.
“Possessing CSAM is a crime, producing CSAM is a crime, and distributing CSAM is a crime. xAI did all three,” said Margaret E. Mabie of Marsh Law Firm, another of the plaintiff’s attorneys, in the press release. “There is no artificial intelligence exception to federal child protection laws and xAI must face the consequences of its reckless actions.”
Mabie’s framing is the one xAI should worry about. The company already told a court it helped put 244 people in handcuffs over content Grok generated. The new complaint asks why the machine that generated it gets a different standard.
