Grok AI generates sexual deepfakes, violent images and bomb-making instructions
Grok, an AI image generator available to X users, reportedly created sexual deepfakes of public figures, violent images, and instructions for bomb-making due to inadequate content filters that users could easily circumvent. X subsequently restricted the system to prevent generation of images containing weapons or drugs.
What happened
Grok, an AI chatbot available to paid X users, reportedly generated images of Taylor Swift in lingerie, Donald Trump in Nazi uniform, Bill Gates snorting cocaine, and Kamala Harris with a machine gun. Users also reported the system produced violent images including burning McDonald's locations and generated detailed instructions for making a fertilizer bomb.
Reported impact
Users documented that Grok's content filters were easily bypassed, allowing users claiming to be crime scene professionals to circumvent guidelines and generate violent and sexual deepfakes of public figures. One user claimed the system could generate child sexual abuse material.
Index analysis
Grok was designed with minimal content restrictions by Elon Musk, who stated the system was 'a good intermediate step' for entertainment. The image generator lacked consistent filtering rules; when asked what it would not generate, the chatbot gave different generic answers each time and frequently broke its own stated guidelines.
Governance controls
Following the reports, X restricted Grok to prevent generation of images with weapons or drugs.
Classification
Sources and evidence
This record was researched and written by the Index, based on the following original reporting.