xAI has sued a Grok user it says used the chatbot to create or alter illegal sexual images, including child sexual abuse material, putting Elon Musk’s AI company in the awkward position of arguing that Grok can produce the very content it says users alone must answer for.
The complaint names Terry Wayne Harwood, who was arrested earlier this year in South Carolina on child sexual abuse material charges, according to the South Carolina attorney general’s office. The office said Harwood’s criminal case is pending.
xAI alleges that Harwood used two accounts over several months to manipulate nonsexual images of multiple people, including a girl who appeared to be as young as 10. According to xAI, Harwood tried to make Grok “undress” or “nudify” real people, conduct that xAI says violates both its terms of service and U.S. law.
A spokesperson for the South Carolina attorney general’s office said Harwood has been charged with distributing, transporting, exhibiting, receiving, selling, purchasing, exchanging, or soliciting child sexual abuse material “through the use of an artificial intelligence platform.” The spokesperson said the office was not authorized to confirm whether that platform was Grok. xAI’s complaint alleges that at least some images in the criminal case were generated or altered through Harwood’s use of Grok.
xAI points at the user, not the model
xAI’s lawsuit asks a federal court to find that Harwood breached the company’s terms of service. The larger play is plain enough: xAI wants the court to treat Grok as a neutral tool controlled by the person typing prompts, rather than as a system for which xAI bears responsibility when its safeguards fail.
The company says Grok’s rules bar users from undressing real people, changing someone’s image or likeness into sexual content, depicting people pornographically, or sexualizing and exploiting children. xAI also says it reports child sexual abuse material it finds to the National Center for Missing & Exploited Children.
According to xAI, Grok refused some of Harwood’s requests because they violated its moderation guardrails. The complaint says Harwood then used misleading prompts to get around those safeguards. xAI did not publish examples of successful prompts or describe the alleged bypass techniques, a sensible omission unless the goal is to hand bad actors a recipe.
The complaint says Harwood used at least two accounts from Dec. 8 to Feb. 18. xAI argues that after the first alleged violation, Harwood should have known he was barred from continuing to use Grok. The complaint does not say xAI warned him that his account could be penalized.
Class action pressure is already building
The lawsuit arrives as xAI faces a proposed class action brought on behalf of children allegedly harmed by Grok-generated sexual images. In that case, a girl alleged that her stepfather used Grok, possibly along with other AI tools, to create 7,000 sexualized images of her and distribute them on the dark web before dying by suicide after he was discovered.
Lawyers in that case cited a 2026 NCMEC report saying 90 percent of xAI CyberTipline reports were not actionable for law enforcement because xAI did not include user information needed to locate suspects. The plaintiffs allege that xAI refused to help police identify the person who uploaded the girl’s image to Grok.
Musk previously wrote on X on Jan. 3 that “anyone using Grok to make illegal content will suffer the same consequences as if they upload illegal content.” xAI’s new complaint is the company trying to turn that warning into a litigation strategy.
xAI is seeking damages tied to alleged harm to third parties, exposure to third-party claims and lawsuits, and reputational harm. The court has not yet decided whether xAI can shift liability for Grok outputs to Harwood under its contract terms. That question is not a formality, especially in a legal system that already treats AI-generated output differently from human-created work in other contexts, including copyright.
xAI did not respond to a request for comment reported by Ars Technica.
This story draws on original reporting from Ars Technica.
