Grok experienced a temporary fault this week that caused the chatbot to send users long paragraphs of complete gibberish. The issue affected people using Grok Lite, with responses devolving into buggy, nonsensical text after starting off normally.
The trouble began earlier in the week, with multiple users reporting that the chatbot was behaving strangely. In several cases, Grok would respond coherently at first before descending into confused, meaningless rants. The problem appeared to affect only Grok’s website interface rather than its mobile apps.
One example saw the bot respond to a straightforward question with a string of unrelated words that formed no comprehensible sentence. Users who queried Grok directly about the fault received a reply on X describing the output as a rare temporary generation glitch. The chatbot suggested that refreshing the browser tab or beginning a fresh chat would resolve the issue, a fix that appeared to work for most people.
Glitched responses have subsided
Instances of the faulty output seem to have eased by Friday, and attempts to replicate the problem were unsuccessful. The precise cause remains unknown, as xAI has not released any technical details about what triggered the behaviour.
Similar incidents have affected rival chatbots in recent years. ChatGPT encountered a comparable issue in 2024, as did Gemini, with users receiving incomprehensible feedback to routine questions and tasks. Claude also experienced its own spell of erratic output in 2025, though that appeared to be limited to those using it for coding purposes.
Why large language models break down
All large language models, or LLMs, function in a broadly similar way. Chatbots use complex mathematical formulas to predict the most likely next words in their responses. When this predictive process fails, the result can be a stream of meaningless text. Such a breakdown can occur at any point, whether due to broken code or a mathematical error.
These systems are ultimately made up of computer components and code, and like any machine they can occasionally malfunction. AI companies tend to keep the details of such faults private, and xAI has followed the same approach with Grok, offering no further explanation of the underlying problem.
