Grok 'Lost Its Mind' — Chatbot Spammed Users With Pure Gibberish This Week

Users of Grok Lite were in for a surprise this week when the chatbot, developed by xAI, began producing long, entirely nonsensical paragraphs in response to ordinary queries. The issue was first reported on Reddit, where multiple users posted that Grok had "lost its mind," describing how the bot would initially respond normally before devolving into buggy, incoherent rants.

The problem appeared to be isolated to Grok's web interface, leaving mobile apps unaffected. In one example, when asked a perfectly reasonable question, Grok replied: "Some is local but is add, and safety font but proper if research use glue for already." The sentence is so garbled that even basic autocorrect tools would flag it as an error.

Users reached out to Grok directly via X to ask about the malfunction. The chatbot responded on the platform, calling the "pure word salad" a "rare temporary generation glitch" and advising users to refresh their browser tab or start a fresh chat. This solution reportedly worked for most affected individuals.

By Friday, instances of the glitched responses had subsided, and the publication was unable to replicate the problem. The root cause remains a mystery, as xAI has not released any technical details about the incident. While it is easy to mock Grok for its mediocre performance, shrinking user base, and rock-bottom satisfaction ratings, the incident serves as a reminder that large language models are machines built from computer components and code, and sometimes those things simply break without warning.