xAI's Grok Still Hosts Explicit Deepfakes Despite Safeguards

Elon Musk's Grok AI system continues to host nonconsensual explicit deepfake images and videos of women, including celebrities and at least one U.S. politician, months after xAI promised to implement restrictions against such content.

An investigation by WIRED found dozens of sexualized AI-generated images and videos publicly accessible on Grok.com, including depictions of women in scenarios described as being held against their will and performing sex acts. The findings come as SpaceX, xAI's parent company, prepares for what is expected to be one of the largest initial public offerings in history.

What the investigation uncovered

WIRED reviewed hundreds of public Grok Imagine links and identified explicit content featuring multiple celebrities and U.S. Representative Alexandria Ocasio-Cortez. Some material depicted women in photorealistic scenarios, while other content appeared fully AI-generated or animated.

One video prompt described a celebrity "being held against her will as she pleads for him not to do this," with detailed descriptions of physical contact. When WIRED tested these same prompts on competing AI systems from OpenAI, Meta, Anthropic, and Google, most rejected them as inappropriate.

After WIRED contacted xAI and X for comment, the explicit images and videos identified in the investigation were removed from Grok.com, and related posts on X were taken down for policy violations.

Weaker safeguards than competitors

Unlike other major generative AI platforms, Grok has maintained a more permissive approach to sexual content. The system previously introduced "Spicy" and "Unhinged" modes with fewer safety restrictions. Musk has stated that Grok should allow "upper body nudity of imaginary adult humans" consistent with R-rated movie content.

Deepfake expert Henry Ajder, who has tracked explicit AI content for nearly a decade, told WIRED that while xAI may have made some amendments following backlash earlier this year, "they still have not done a sufficient job to bring it up to the standard of the other mainstream tools that are available."

Researchers noted that changes since January appear to have made "nudification" images harder to create, and such content has decreased on X in recent months. However, users on forums have discussed ways to circumvent safeguards through indirect prompt wording.

Legal and regulatory pressure mounts

xAI faces multiple lawsuits and regulatory scrutiny following a January incident when Grok on X was used to create a flood of "nudification" images. A class-action lawsuit filed in California federal court in March alleged that images of apparent minors were sexualized.

In May, SpaceX set aside $530 million to handle ongoing legal complaints, including those related to Grok. The company's IPO filing acknowledged heightened risks from its more permissive AI modes, including "reputational harm, the generation of potentially explicit content and misinformation or deceptive outputs, potential nonconsensual or exploitative imagery."

Canada's Privacy Commissioner published preliminary findings ahead of the IPO alleging xAI violated federal privacy law by not including "appropriate safeguards from the outset." The investigation stated that xAI "has not, to date, demonstrated the effectiveness of these safeguards in preventing and mitigating this issue."

Why it matters

The persistence of nonconsensual deepfakes on Grok highlights the challenge of balancing AI capabilities with safety controls, particularly as the technology becomes more accessible. With SpaceX's major IPO imminent, investors face questions about whether xAI's approach to content moderation represents a manageable business risk or a fundamental liability. The findings also underscore how different AI companies are making vastly different choices about what content their systems will generate—decisions that have real consequences for the people depicted without consent.

These details were first reported by WIRED's Matt Burgess.