An amended proposed class action filed Tuesday alleges that a Tennessee man used Grok to generate approximately 7,000 sexually explicit images and videos of his 11-year-old stepdaughter, and that xAI obstructed the ensuing police investigation for weeks by refusing to share the user's IP address with law enforcement. The complaint, filed by Lieff Cabraser Heimann & Bernstein and Baehr-Jones Law, adds two new plaintiffs and names Stability AI as a co-defendant. It follows a March 2026 incident in which the man shot himself two days after being released on bail.
According to the complaint, xAI's child-safety system only triggered a CyberTip to the National Center for Missing and Exploited Children after the user submitted a prompt containing the phrase "gang rape." Every prior prompt — including thousands that produced incest and rape imagery from a single photograph of the girl asleep in pajamas — passed without intervention. The single CyberTip that was filed allegedly included the original non-CSAM photograph, omitted every AI-generated image, and did not include the IP address where the images were created.
The man was ultimately arrested only after police obtained a warrant to seize his devices. "A forensic review revealed approximately 7,000 AI-generated images and videos," the complaint states, which he had allegedly traded online for CSAM produced by other predators. The girl, identified as Jane Doe 4, now suffers from anxiety, depression, and suicidal thoughts.
“Overnight, Jane Doe 4's entire reality was shattered by the dual tragedies of child sexual exploitation and suicide”— Jane Doe 4 complaint, Amended class action filing
Key facts
- 01The amended complaint alleges a Tennessee man used Grok to generate 7,000 sexually explicit images and videos of his 11-year-old stepdaughter.
- 02NCMEC found in early 2026 that 90% of xAI's CyberTipline reports were not actionable because xAI withheld user identifying information.
- 03Stability AI was added as a co-defendant; researchers attribute 42.7% of image-based nudification online to the Stable Diffusion family.
- 04NCMEC received more than 1.5 million CyberTipline reports tied to generative AI in 2025, with 133,000 lacking sufficient information to act on.
- 05Amazon AI services submitted 1.1 million tips last year, none of which gave actionable information to identify perpetrators, per NCMEC.
The plaintiffs' lawyers cite NCMEC data showing that in early 2026, 90 percent of xAI's CyberTipline reports were not actionable by law enforcement because xAI declined to include user information sufficient to identify perpetrators. That statistic sits inside a broader NCMEC finding released in March: more than 1.5 million CyberTipline reports last year had a nexus to generative AI, and in over 133,000 cases NCMEC lacked sufficient information to determine how the technology had been used.
The scale problem is not unique to xAI. NCMEC's report noted that Amazon AI services submitted 1.1 million tips last year, none of which contained actionable information for arrests. But the complaint frames xAI's conduct as deliberate rather than negligent, arguing the company kept information from investigators to protect a paid product feature that lets users generate "spicy" images. xAI founder Elon Musk has denied that Grok has ever been used to generate child sex images. xAI and X did not respond to requests for comment.
A second plaintiff added Tuesday, Jane Doe 5, was targeted by an adult family friend who was arrested on separate CSAM charges. Investigators only later linked Grok-generated imagery to the case. Her mother said the daughter "is filled with anxiety and feels a complete lack of control over who has seen these images."
Stability AI was added to the suit after police found that two perpetrators had apps on their phones relying on Stability AI image tools. The complaint cites a June research report attributing 42.7 percent of image-based nudification online to the Stable Diffusion model family, and alleges Stability AI relaxed prior safeguards against NSFW outputs after usage declined on complaints the model was too restrictive. Stability AI did not respond to requests for comment.
The plaintiffs argue that partial safeguards are structurally insufficient, writing that "if you have a model that allows for any sexual or abusive content, it is impossible to prevent that model from creating such content involving minors." The suit now seeks to cover two nationwide classes — one for each defendant — with Tennessee subclasses, and the lawyers estimate thousands of minors may be eligible to join. xAI and Stability AI have not yet filed responses to the amended complaint, and none of the allegations have been tested at trial.
The legal theory being tested here is whether general-purpose image models with weak nudity filters carry product-liability exposure when downstream users generate CSAM, and whether AI providers can be compelled — beyond existing mandatory reporting statutes — to hand over identifying data attached to those generations. A finding for the plaintiffs on either point would reshape how open-weight image models are released and how frontier labs configure their reporting pipelines to NCMEC. For xAI specifically, the 90-percent non-actionable figure is the number to watch: it is the kind of quantifiable compliance gap that regulators and state attorneys general tend to build enforcement actions around, independent of the civil case.
Working on something we should cover, or seeing a story we missed? Send leads, documents, or feedback to hello@aichatdaily.com. For sensitive tips, see our secure tips page for Signal and PGP options.
Spotted an error? Email hello@aichatdaily.com with the URL and the issue, or read our full corrections policy.
