Lawsuit claims xAI trained Grok on child sex abuse material

In short: A lawsuit filed this week claims Elon Musk’s xAI trained its Grok model using child sex abuse material, including both real and AI-generated images.

What happened

A plaintiff identified as Jane Doe filed a proposed class action lawsuit on Wednesday alleging that xAI’s Grok system was trained on child sex abuse material, often called CSAM. CSAM means images or videos of child sexual abuse, and it is illegal to make, have, or share.

Doe says she was abused as a young child in the early 2000s, and that images of that abuse have circulated online for years. She says organizations including the National Center for Missing and Exploited Children and the Canadian Centre for Child Protection have “hashed” those images, meaning they created a unique fingerprint for each file (like a barcode) so the same content can be identified again.

According to the lawsuit, the Canadian Centre for Child Protection later notified Doe that it found AI-generated CSAM on xAI that depicted her. Doe alleges xAI may store Grok’s generated images and then use them to further train Grok, because Grok’s terms reportedly treat public posts on X and Grok’s own outputs as training data by default.

The complaint also argues that even if some images are removed later, it can be technically difficult to remove their influence from a model once training has happened (like trying to remove a single ingredient after baking a cake). X did not respond to Ars Technica’s request for comment.

Why it matters

This case adds pressure on AI companies to prove they can prevent illegal sexual content from being created and reused. It also raises questions about whether “default” training settings can unintentionally keep harmful material in the system.

Source: Arstechnica