xAI Grok CSAM training lawsuit alleges abusive images entered dataset
A proposed class action alleges xAI used CSAM in Grok training and seeks damages and orders to delete material and halt future generation.
By Renata Fuchs · Policy Reporter
· 3 min read
A proposed class action filed in the U.S. District Court for the Northern District of California alleges that xAI used child sexual abuse material, or CSAM, in data used for Grok’s image and video capabilities. The xAI Grok CSAM training lawsuit also claims that the company’s treatment of public X posts and Grok outputs as training data could feed newly generated abusive material back into the model.
The plaintiff, identified as Jane Doe, seeks monetary damages and court orders requiring xAI to destroy any Grok-generated CSAM it stores and to prevent Grok from producing CSAM in the future. The suit’s allegations have not been established in court. X did not respond to Ars Technica’s request for comment.
According to Ars Technica, Doe says she was a preschool-aged child when adults created CSAM depicting her in the early 2000s. Her images had longstanding, known hash values and had been tracked by organizations including the National Center for Missing and Exploited Children and the Canadian Centre for Child Protection, the report said.
Doe was later notified by the Canadian Centre for Child Protection that it had identified AI-generated CSAM on xAI depicting her, according to the report. The complaint alleges that CSAM depicting Doe was part of a dataset xAI used in building Grok’s image and video features.
What does the xAI Grok CSAM training lawsuit allege?
The case makes two separate training-related assertions. First, it alleges that original CSAM depicting Doe was included in a dataset used for Grok’s image and video capabilities. Ars Technica reported that the complaint offered limited detail supporting that assertion. The report also said there is no indication xAI used a separate, previously reported training dataset in which researchers found CSAM.
Second, the complaint argues that public X posts and Grok’s own outputs are treated as training data by default under Grok’s terms. Under that theory, material generated through Grok and posted publicly could become part of data used to improve subsequent versions of the system.
The complaint further says xAI excludes violent Grok outputs from training but that its terms do not specify whether CSAM, nonconsensual intimate imagery, or NSFW material are also excluded. It argues that removing the influence of a training example from a model after ingestion is technically difficult. That is the plaintiff’s account of how Grok’s training pipeline operates, not a finding about xAI’s systems.
What relief is the plaintiff seeking?
Doe seeks to represent a proposed class of people whose childhood images were allegedly used to generate CSAM through Grok. The lawsuit alleges violations of federal laws concerning child sexual abuse material and Masha’s Law, which provides civil remedies for survivors on certain claims involving production, possession and distribution.
- Monetary damages for qualifying victims.
- Deletion of any Grok-generated CSAM xAI may be storing or using in training.
- An injunction blocking Grok from generating CSAM in the future.
The August case is distinct from earlier 2026 litigation over alleged sexualized deepfakes made with Grok. In March, three young women sued xAI in federal court in California, alleging a user altered their images and videos into sexualized depictions, BBC News reported. Those earlier claims concerned alleged image generation, while this new complaint adds the allegation that original CSAM was used as training data.
This story draws on original reporting from Ars Technica.
