The chatbot Grok, under xAI, recently experienced an abnormal failure, with some users encountering continuous garbled and meaningless text when using Grok.com. Affected users reported that after requesting Grok to generate a PDF, the model continuously output multiple incomprehensible sections, and some responses even included source links pointing to websites related to reinforcement learning research.
According to reports, affected users mainly used Grok Lite, and the issue was first discovered on Wednesday morning. Testing failed to reproduce the problem, indicating that the fault may only affect a small number of users. Some users reported that refreshing or restarting the session usually restored normal function, but others continued to receive garbled text even after multiple refreshes. The issue seems to currently only appear in direct queries on Grok.com, while Grok on X.com is not affected.
The official Grok account later confirmed that this was a "rare temporary generation failure" and stated that restarting the chat or regenerating the conversation usually restores normal function immediately. The official status page shows that all Grok services are running normally, with no recorded service incidents.
This incident occurs as xAI is currently experiencing high personnel turnover. According to The Information's report in May this year, xAI has lost most of its founding team members and at least 50 researchers and engineers over the past few months. In July, xAI released its latest foundation model, Grok4.5, claiming it reached the "Opus level," while also offering faster speed, higher token efficiency, and lower cost. This anomaly once again highlights that stability in the reasoning, generation, and retrieval chains of large models remains a critical challenge for large-scale deployment.
