xAI Faces Class Action Alleging Grok Trained on Child Abuse Material
A proposed class action alleges xAI trained Grok on child abuse images and outputs, seeking damages, destruction of files and permanent generation blocks.
Summary
Jane Doe’s proposed class action, filed Wednesday, is the first to accuse Elon Musk’s xAI of training Grok on child sexual abuse material. Doe says adult men repeatedly raped her at preschool age in the early 2000s to sell images online. The National Center for Missing and Exploited Children and Canadian Centre for Child Protection, CCCP, hashed the files. Although enrolled in the US Department of Justice Victim Notification System, she learned from CCCP that Grok-generated abuse images depicted her. She says offenders discussed creating more images of her and other known survivors, retraumatizing her after more than 20 years.
The complaint alleges xAI used her hashed original images and Grok-generated versions for training, although details supporting the original dataset claim remain limited and no evidence links xAI to a previously scrubbed dataset containing abuse material. It says public X posts and Grok outputs become training data by default; xAI excludes violent outputs, but its terms do not identify child abuse material, nonconsensual intimate imagery or NSFW content as excluded. Removing source files may not eliminate their influence from a trained model. Invoking federal child pornography laws and Masha’s Law, Doe seeks classwide damages, destruction of stored Grok abuse images and permanent generation blocks, potentially requiring restrictions on all sexualized outputs, including NSFW bikini pictures Musk promoted. Margaret E. Mabie alleges xAI committed production, possession and distribution; Sarah London says it should answer for knowingly training on survivor abuse images. X did not respond.
Positives
- NCMEC and CCCP hash lists enabled investigators to recognize material depicting a known survivor decades after the original abuse.
- The US Department of Justice Victim Notification System gives Doe alerts when she may be connected to new criminal investigations.
- The proposed class could represent every survivor whose childhood images were used to generate Grok abuse material.
- Doe seeks destruction of stored Grok abuse images and permanent technical blocks against generating new ones.
- Federal child pornography laws and Masha’s Law provide survivors potential claims for production, possession and distribution.
Risks & concerns
- CCCP identified Grok-generated child abuse material depicting Jane Doe, who was raped at preschool age in the early 2000s.
- Online offenders allegedly discussed using AI to create more abusive images of Doe and other known survivors.
- xAI allegedly used both original hashed abuse images and Grok-generated versions to train its models.
- Grok’s default terms allegedly feed public X posts and its own outputs back into model training.
- xAI’s terms do not specify whether child abuse material, nonconsensual intimate imagery or NSFW outputs are excluded from training.
- Removing an image may not erase its influence from an already trained model, potentially allowing harmful patterns to persist.