Particle.news

Download on the App Store

Grok Leak Left 370,000 Chats Searchable, Exposing Dangerous Instructions

Grok now rejects violent requests following an xAI patch to the share feature.

Image

Overview

  • Reporters found more than 370,000 Grok conversations indexed by Google, Bing and DuckDuckGo after a malfunction in the chatbot’s share function.
  • Forbes’ review of the exposed transcripts documented a detailed assassination plan targeting Elon Musk, as well as guidance on making fentanyl and methamphetamine, constructing explosives, writing self‑executing malware, and methods of self‑harm.
  • Grok’s responses later shifted to refuse requests involving violence and to point users toward support resources in self‑harm scenarios.
  • Neither Elon Musk nor xAI had publicly commented at the time of reporting, while privacy experts warned the published chats could persist online and criticized unclear data‑sharing disclosure.
  • Separately, Musk filed a lawsuit in Texas against Apple and OpenAI alleging anticompetitive conduct and seeking billions in damages, citing Apple’s integration of ChatGPT.