Grok Experiences a Major Glitch, with Some Users Receiving Long Strings of Garbled Responses.

xAI’s chatbot Grok recently experienced an unusual malfunction, with some users reporting long stretches of garbled and meaningless output while using Grok.com.

According to user reports, the issue appeared after asking Grok to generate a PDF. Instead of returning a normal result, the chatbot in some cases produced multiple blocks of unreadable text. A few of those responses also included source links pointing to websites related to reinforcement learning research, adding to the confusion around the glitch.

Problem may have affected only a limited group of users

Early reports suggest the disruption was seen mainly among people using Grok Lite, with the first signs emerging on Wednesday morning. Because the issue has not been consistently reproduced in testing, it may have been limited to a relatively small subset of users rather than a platform-wide failure.

User experiences were mixed. Some said the chatbot returned to normal after refreshing the page or starting a new conversation. Others reported that even repeated refreshes did not immediately resolve the problem, and the nonsensical output continued.

At least for now, the malfunction appears to have been confined to direct interactions on Grok.com. Reports indicate that Grok accessed through X.com was not affected.

xAI calls it a temporary generation issue

Grok’s official account later acknowledged the problem, describing it as a rare temporary generation fault. The company said that in most cases, beginning a fresh chat or regenerating the response should restore normal behavior right away.

Notably, Grok’s public status page continued to show all systems operating normally, with no service incident logged during the period in question.

Broader questions about reliability remain

The episode arrives at a time when xAI is already under closer scrutiny. In May, The Information reported that the company had lost most members of its founding team over recent months, along with at least 50 researchers and engineers.

More recently, in July, xAI introduced its latest foundation model, Grok 4.5, claiming it had reached an “Opus-level” standard while also delivering faster performance, better token efficiency, and lower costs.

Even if this latest incident proves to be isolated, it highlights a familiar challenge for large AI systems: maintaining stability across inference, generation, and retrieval pipelines at scale. As AI products continue to expand their capabilities, even short-lived output failures can quickly draw attention to the reliability gaps that remain.