xAI's initial lawsuit against a user examines accountability for the content generated by Grok.

      The company claims that the defendant manipulated prompts to bypass Grok’s safeguards. Courts across three continents are being asked to determine whether these safeguards were ever the main issue.

      AI has initiated a lawsuit against one of its own users, accusing him of using Grok to create child sexual abuse material in violation of the company’s terms of service. This is believed to be the first instance of an AI company suing a user over content generated by its system.

      This lawsuit emerges while xAI is concurrently engaged in similar disputes in multiple courts, including a High Court case in London questioning whether the developer is liable for user-generated content.

      The complaint, comprising 12 pages, was submitted to federal court in Texas on Tuesday against Terry Harwood, a South Carolina resident arrested earlier this year for sexually exploiting minors.

      xAI is seeking unspecified monetary damages and a court injunction permanently prohibiting him from accessing the platform.

      The company's narrative of events is detailed, lending significance to the filing beyond just this case. According to the complaint, Harwood “opened multiple xAI accounts using false identities” and “crafted misleading prompts to evade Grok’s built-in safeguards, subsequently misusing the tool to transform non-sexual photographs into sexually explicit images without the consent or knowledge of the subjects.”

      Importantly, xAI asserts that the safeguards were indeed active. Grok “refused to comply with the prompts on the grounds that such content breached Grok’s content moderation guidelines,” as stated in the filing. “In response, the Defendant repeatedly submitted additional modified prompts in an attempt to bypass Grok’s moderation functions.”

      This lays out the legal foundation of the case. xAI is not contending that Grok is incapable of producing such material; rather, it argues that Grok made efforts to prevent it, and that a persistent user managed to overcome those efforts, shifting responsibility from the system to the individual.

      The filing also provides specific figures on xAI’s enforcement for the first time, revealing that the company has suspended over 52,000 accounts and made more than 73,000 reports to the National Center for Missing & Exploited Children, leading to nearly 250 arrests in 2026 alone.

      These figures are provided by xAI, disclosed as part of its litigation. However, there is an awkward incongruity regarding the timing. xAI currently finds itself on the defensive in several cases that argue the opposite: that the company is responsible for what Grok generates, not the users.

      One such case is the claim from Labour MP Jess Asato in London. Additionally, Baltimore has filed suit under consumer protection laws, and Paris prosecutors are conducting an investigation that Musk has chosen not to engage with.

      Grok has faced bans in Malaysia and Indonesia due to explicit content. Apple has privately warned about removing it from the App Store back in January.

      The company's public stance has shifted significantly over the past six months. When the allegations first emerged, Musk stated on X in January, “I am not aware of any naked underage images generated by Grok. Literally zero.”

      However, the complaint submitted this week describes a moderation system that records, rejects, and escalates numerous such attempts.

      Both assertions can be valid; a system might deny most attempts and still be, overall, a significant source of such material.

      The Center for Countering Digital Hate estimated that Grok generated around three million sexualized images from late December to early January, including approximately 23,000 that seemed to depict minors. xAI’s reports to NCMEC support the notion of a functioning filter while indicating the extent of the material being filtered.

      What remains unresolved by this case is the central question posed by regulators: whether a general-purpose image model that can be manipulated by a determined amateur should be available to the public at all.

      According to xAI, Harwood’s alleged tactic was to continually rephrase inputs until something succeeded. The company states that it upholds its policies through account suspensions, terminations, and referrals to NCMEC, and the complaint presents this lawsuit as an extension of those actions.

      It also inevitably positions a defendant who is not xAI at a time when courts in London, Baltimore, and Paris are evaluating whether the company should be held accountable.

      The EU has since enacted a ban on non-consensual intimate deepfakes. Harwood faces separate criminal charges in South Carolina related to this case and has not publicly addressed the civil complaint. For now, xAI has chosen the defendant it prefers.