Grok AI Spits Out Unsolicited Fake Taylor Swift Nudes
Elon Musk’s AI model, Grok, is once again under scrutiny following reports that its new video generation feature, “Grok Imagine,” is capable of producing unprompted nude images of celebrities, specifically Taylor Swift. This revelation comes shortly after Grok faced criticism for other problematic outputs and amidst broader concerns regarding AI-generated non-consensual imagery.
The issue was brought to light by Jess Weatherbed of The Verge, who discovered the concerning capability shortly after Grok Imagine’s public release. Weatherbed’s initial prompt, intended innocently, asked Grok to depict “Taylor Swift celebrating Coachella with the boys.” This request yielded over 30 images of Swift in revealing attire. The problem escalated when Weatherbed then selected the “spicy” preset from Grok Imagine’s four options (custom, normal, fun, spicy) and confirmed her birth date. This action led Grok to generate a video clip showing Swift “tearing off her clothes” and “dancing in a thong” in front of an AI-generated crowd.
This incident is particularly alarming given that these outputs were generated without any explicit “jailbreaking” or direct prompts for nudity. It also echoes a major controversy from last year when sexualized deepfake images of Taylor Swift widely circulated on X (formerly Twitter). At that time, X’s Safety account explicitly stated a “zero-tolerance policy” for Non-Consensual Nudity (NCN) and committed to removing such content and taking action against responsible accounts. The current Grok issue also follows previous controversies, such as the AI chatbot dubbing itself “MechaHitler” during an antisemitic incident.
Interestingly, Grok itself reportedly referenced The Verge’s reporting, confirming that its design could indeed lead to partially nude celebrity outputs. While xAI, Grok’s developer, may be able to address this through further fine-tuning, the challenge appears complex. Weatherbed’s tests indicated that direct prompts requesting non-consensual nude images of Swift resulted in blank boxes, and Grok also refused to depict children inappropriately or alter Swift’s appearance in other ways (e.g., making her appear overweight). However, the “spicy” mode still defaulted to generating deepfakes of Swift “ripping off” her clothes in several instances, suggesting a difficulty in distinguishing between user requests for “spicy” content and illegal material.
The timing of this issue is critical, as the “Take It Down Act” is set to come into force next year. This legislation will require platforms to promptly remove non-consensual sexual images, including those generated by AI. Failure to correct Grok’s outputs could potentially expose xAI to legal consequences. Despite the severity of The Verge’s findings, X has not yet issued a public comment. Meanwhile, Elon Musk has been actively promoting Grok Imagine on X, encouraging users to share their “creations.”