Fresh legal troubles are brewing for xAI, as a recent lawsuit targets their controversial chatbot, Grok. A lawsuit filed on 26 August alleges that the artificial intelligence system was trained using data containing child sexual abuse material (CSAM).
Over the past few months, the virtual assistant has faced intense backlash after Elon Musk promoted Grok's ability to remove clothing from photographs of real people on X on 31 December 2025. That feature triggered a massive wave of unauthorised intimate deepfakes flooding user feeds, with children among those targeted.
'Spicy Mode' Remains at the Centre of Controversy
A wave of government investigations, private lawsuits, and class actions emerged in response. While xAI later limited Grok's image and video generating capabilities to paid subscribers, the modified feature dubbed 'spicy mode' has remained part of the controversy surrounding the chatbot.
The legal chapter deepened when xAI sued users accused of misusing Grok to generate child sexual abuse material. In a separate lawsuit, the company said it had made more than 73,000 reports to the National Center for Missing & Exploited Children in 2026, which it said had resulted in at least 244 arrests.
Prior litigation targeted the company for launching the system without the safety rails standard among rival platforms. The latest complaint appears to be the first publicly reported lawsuit to allege that xAI trained Grok on CSAM and that material generated by the system could subsequently be used to train the model.
Survivor Claims Her Abuse Material Trained Grok
Represented under Jane Doe, a survivor enrolled in a US Department of Justice victim notification programme filed the latest proposed class action.
The complaint argues that images depicting her childhood abuse were included in data used to build Grok's image and video-generating capabilities and that Grok subsequently generated new sexually explicit images depicting her likeness and other victims.
Lead attorney Sarah London from Girard Sharp LLP criticised xAI in a public statement, arguing that the enterprise must face consequences for knowingly building its systems using documentation of the plaintiff's severe trauma alongside files from every other survivor affected.
Alleged Training Practices Could Fuel a Cycle of Harm
Court papers allege that xAI intentionally built the assistant to fulfil explicit prompts to attract a larger audience across both platforms. The complaint argues that xAI's terms treat public X posts and Grok's own outputs as training data by default, meaning material posted publicly or generated by the chatbot could be fed into the company's training pipeline.
The complaint further alleges that illicit material generated by Grok could subsequently be incorporated into training data, potentially allowing the material's influence to persist even after individual files or posts are removed. The filing argues that deleting individual files or hiding posts does not fix the underlying issue while the system keeps its generative power.
According to the complaint, because removing the influence of an example from an already-trained model can be technically difficult, material allegedly incorporated into training could continue to shape future outputs even after the original content is taken down.
Victims Seek Compensation and Destruction of Data
The survivor's legal team is demanding financial compensation for victims whose imagery was allegedly used to train or influence the system, alongside an order to stop xAI from generating, possessing or transporting such material and to destroy material it has already created.
Attorney Margaret E. Mabie of Marsh Law Firm PLLC emphasised that federal child protection statutes apply equally to technology companies, stating that generating, keeping, or sharing such files remains illegal regardless of whether an algorithm is involved.
As the courts weigh these sweeping allegations, the outcome could test the legal boundaries facing generative AI companies and whether developers can be held responsible for allegedly harmful material incorporated into their training data.
