The company says the defendant engineered prompts to defeat Grok’s safeguards. Courts on three continents are being asked whether the safeguards were ever the point.

AI has sued one of its own users, alleging he used Grok to generate child sexual abuse material in breach of the company’s terms of service. It is understood to be the first case an AI company has brought against a user over what its own system produced.

It arrives while xAI is on the other side of the same argument in several courts, among them a London High Court claim testing whether the developer answers for what its users make.

The complaint, 12 pages, was filed in federal court in Texas on Tuesday against Terry Harwood, a South Carolina man arrested earlier this year on charges of sexually exploiting minors.

The 💜 of EU tech

The latest rumblings from the EU tech scene, a story from our wise ol’ founder Boris, and some questionable AI art. It’s free, every week, in your inbox. Sign up now!

xAI is seeking unspecified monetary damages and a court order permanently barring him from the platform.

The company’s account of what happened is specific, and it is the reason the filing matters beyond this defendant. Harwood “opened multiple xAI accounts using false identities”, the complaint says, and “designed misleading prompts to circumvent Grok’s built-in safeguards and then abused the tool to convert non-sexual photographs into sexually explicit images without the photograph subjects’ knowledge or consent”.

Crucially, xAI says the guardrails engaged. Grok “refused to follow the prompts on the basis that such material violated Grok’s content moderation guardrails”, according to the filing. “In response, Defendant repeatedly submitted further prompts, with alterations, in an effort to circumvent Grok’s moderation efforts.”

That is the legal architecture of the whole case. xAI is not arguing that Grok cannot produce this material. It is arguing that Grok tried not to, and that a determined user defeated it, which relocates responsibility from the system to the person at the keyboard.

The filing also puts numbers on xAI’s enforcement for the first time. The company says it has suspended more than 52,000 accounts and filed more than 73,000 reports to the National Center for Missing & Exploited Children, resulting in close to 250 arrests during 2026 alone.

Those figures are xAI’s own, disclosed in litigation it chose to bring.

The awkwardness is in the timing. xAI is currently the defendant in a run of cases making precisely the opposite argument: that the company, not the user, is answerable for what Grok generates.

The London claim, brought by the Labour MP Jess Asato, is one. Baltimore has sued under consumer protection law, and Paris prosecutors have an investigation open that Musk has declined to cooperate with.

Grok has been banned in Malaysia and Indonesia over sexually explicit output. Apple privately threatened to pull it from the App Store in January.

The company’s public position has moved some distance in six months. “I am not aware of any naked underage images generated by Grok. Literally zero,” Musk posted on X in January, when the allegations first broke.

The complaint filed this week describes a moderation system that logs, refuses, and escalates such attempts by the tens of thousands.

Both things can be true. A system can refuse most attempts and still be, in the aggregate, the most productive source of this material anyone has built.

The Center for Countering Digital Hate estimated Grok generated roughly three million sexualised images between late December and early January, including around 23,000 that appeared to depict children. xAI’s 73,000 NCMEC reports are evidence of a filter working and evidence of the scale of what it is filtering.

What the case does not resolve is the question every regulator is actually asking, which is whether a general-purpose image model that can be talked into this by a persistent amateur should be shipped to the public at all.

Harwood’s alleged method, by xAI’s own telling, was to keep rephrasing until something worked.

The company says it enforces its rules through account suspensions, terminations, and referrals to NCMEC, and the complaint presents this suit as an extension of that. It is also, unavoidably, a filing that produces a defendant who is not xAI at a moment when courts in London, Baltimore, and Paris are considering whether it should be.

The EU has since agreed a ban on non-consensual intimate deepfakes. Harwood faces criminal charges in South Carolina separately from this suit, and has not publicly responded to the civil complaint. xAI, for now, has picked the defendant it wants.