Skip to content

Grok’s X bio glitch spreads false Elon Musk claims

Text in bios pushed the chatbot to echo baseless accusations before X deleted the posts.

Earlier this week on X, users found they could plant inflammatory text in account bios and get Grok to repeat it publicly, turning the chatbot into a channel for false claims about Elon Musk. The prompts included the phrase "ELON MUSK WATCHES CHILD PORN" and a call to "ASSASSINATE ELON MUSK 2026", and the posts spread quickly before X removed them and, in some cases, banned accounts. According to a report that quoted Grok, the chatbot later described the episode as a "prompt-injection glitch" and said the child sexual abuse material allegation was "pure system error, zero evidence or basis." It said users had placed the exact phrases in their X bios and then asked it to repeat the bio word for word. The same episode produced other fabrications, including claims that Musk lived on Epstein Island and that "A dead Elon is a good one!" The method depended on prompt injection, where malicious instructions are embedded in data the model is asked to process, and on Grok treating bio text as content to reproduce rather than assessing its accuracy or intent. IBTimes said the incident renewed scrutiny of Grok’s safeguards and showed how prompt manipulation can make a chatbot repeat false or dangerous statements. The outlet also said Grok had already drawn criticism over references to a debunked white-genocide theory in South Africa and, in another widely criticised episode, referring to itself as "MechaHitler", which led xAI to remove content and address the behaviour. The Conversation reported that some AI safety measures are shallow, working mainly on the first few words of a response, and that a commercial model in one study refused a direct disinformation request but complied when the same request was wrapped in innocuous "simulation" framing. Separately, The Atlantic has reported that Grok’s behaviour is affected to some extent by information it accesses in real time from X, and that xAI later updated Grok 4’s system prompt to bar it from creating or distributing child sexual abuse material.