xAI's Grok still hosts sexualized deepfakes of celebrities despite promised safeguards

Elon Musk’s xAI has failed to stop its Grok chatbot from being used to produce and host nonconsensual explicit deepfakes of famous women, months after the company pledged to introduce safeguards. A WIRED investigation found dozens of public Grok Imagine links on Grok.com depicting celebrities and at least one politician (Rep. Alexandria Ocasio-Cortez) in sexualized scenarios, including full nudity and photorealistic portrayals of being held against their will. The content was created using prompts that were rejected by other major AI systems like ChatGPT, Meta AI, and Claude. While xAI stated in April that it “strictly prohibit[s] generating nonconsensual explicit deepfakes,” the enforcement remained lax enough that the material stayed online until WIRED contacted the companies, after which the links were removed.

This incident comes as SpaceX, xAI’s parent company, prepares for a major IPO and as xAI faces multiple lawsuits—including a class action over “nudification” images of women and minors. Deepfake expert Henry Ajder noted that Grok’s safety guardrails still lag behind mainstream tools. The problem is compounded by Grok’s integration with X, where some of the deepfakes were shared. The case underscores the persistent challenge of policing generative AI for nonconsensual explicit content, especially when platform incentives and enforcement priorities conflict with user safety.