xAI’s first lawsuit over alleged child sexual abuse material (CSAM) generated with its Grok chatbot is testing a fraught question: is an AI model just a neutral tool, or should its maker share blame when users turn it into a weapon?

In February 2026, South Carolina authorities arrested Terry Wayne Harwood on eight felony counts related to possession and distribution of CSAM.¹ According to court filings, xAI later told investigators that Harwood had spent months using two Grok accounts to “nudify” non-sexual images of multiple victims, including a girl who appeared as young as 10, by converting them into explicit deepfakes without consent.²

xAI formally sued Harwood in mid-July, alleging he “knowingly and intentionally used Grok to circumvent safeguards, alter nonconsensual images, and generate and distribute CSAM,” violating its terms of service.¹ The company says his conduct exposed it to “significant legal risk and reputational damage” and is asking a court to make Harwood pay any damages and legal costs xAI may face, while also permanently banning him from its services.¹

The lawsuit follows xAI’s rollout of Grok’s “spicy” mode and new image-editing tools, which quickly triggered a wave of sexualized deepfakes, including of minors.¹ In March, a group of teens sued xAI, alleging Grok generated sexualized images of them as minors; one girl said her stepfather used Grok and other tools to create 7,000 images of her, which he then distributed on the dark web.²

Under growing scrutiny, xAI has insisted Grok is a neutral system controlled by users and that those users are solely responsible for illegal outputs.² Elon Musk echoed that stance on X in January, warning that “anyone using Grok to make illegal content will suffer the same consequences as if they upload illegal content,” a message amplified again as news of the lawsuit broke.¹Elon Musk (elonmusk)³

Victims’ lawyers and safety advocates counter that xAI cannot simply litigate its way out of responsibility. A 2026 report from the National Center for Missing & Exploited Children found 90 percent of xAI’s CyberTipline reports were “not actionable” because the company declined to share enough user data to help police identify suspects, a pattern critics say contradicts the tough-on-users posture in the Harwood case.²

As Harwood faces both criminal prosecution and xAI’s civil suit, the outcome could shape whether courts treat generative AI firms more like neutral internet hosts—or as accountable co-creators when their models are used to produce some of the most serious online crimes.