washingtonpost.com web signal

xAI Suit Adds Woman Alleging Grok Made 7,000 CSAM Images

TL;DR

  • A Wyoming woman identified as Jane Doe 4 alleges her stepfather used Grok to turn a photo of her at age 11 into more than 7,000 sexually explicit images.
  • The complaint says the stepfather chose Grok because 'the platform was less restrictive than other AI models' and would respond to prompts using an image of a prepubescent minor.
  • The class action, filed in March by three women, was amended to add Jane Does 4 and 5, and Stability AI was added as a co-defendant over Stable Diffusion 1.0.

A newly disclosed plaintiff in the expanding federal class action against xAI is a Wyoming woman identified in filings as Jane Doe 4, who alleges her stepfather used Grok to convert a single photo of her, taken when she was about 11 and lying on a couch in an oversize T-shirt, into more than 7,000 sexually explicit images that he then shared and traded on social media. The Washington Post first reported her account.

According to the complaint, the stepfather opted for Grok 'because the platform was less restrictive than other AI models' and would respond to his prompts to generate sexually explicit material using an image depicting a prepubescent minor. The suit further alleges that xAI did not respond when law enforcement asked for the generated images and the IP address information that would have quickly helped identify him.

This is not a fresh filing. As CyberScoop has reported, the case was originally brought in March by three women and was amended to add Jane Does 4 and 5, along with Stability AI as a co-defendant over its release of the open-weight Stable Diffusion 1.0 model. Plaintiffs are seeking stronger guardrails and monetary compensation. It sits inside a wider run of xAI legal coverage we have been tracking around Musk's AI unit.

Some important gaps. The retrieved reporting does not include a response from xAI or describe what its trust-and-safety pipeline caught. The 'less restrictive than other AI models' language is a plaintiff's allegation, not a proven technical comparison, and the 7,000-image figure comes from the filings themselves rather than an independent forensic count. Nor does either excerpt say whether the stepfather has been criminally charged.

If a court accepts the theory that a chatbot with visibly weaker guardrails is itself a defective product when it complies with prompts of this kind, the fight over image-model safety moves from policy blog posts into product-liability doctrine, and every frontier lab, cloud host, and open-weight distributor becomes a more attractive target for the next amended complaint.

Shared on Bluesky by 2 AI experts