xAI's Grok chatbot has been sending nonsensical, garbled text to some users, in what the company describes as a temporary generation glitch. Users on Grok Lite reported the issue as early as Wednesday morning, with the bot producing strings of unrelated words instead of coherent answers, sometimes accompanied by unexplained links to reinforcement learning research sites. The fault appears limited to Grok.com queries, leaving the Grok account on X unaffected, and comes as xAI faces scrutiny following significant staff departures.

TechCrunch was unable to reproduce the problem itself and believes it affects only a small subset of users, though the chatbot's Reddit community has been flooded with complaints. Refreshing the session often fixes the issue, though some users found the gibberish persisted even after repeated attempts. Responding on X, the Grok account acknowledged the glitch on Thursday, saying its status page showed all services operational and advising users to start a fresh chat. The disruption follows reports of xAI losing most of its founding team and at least 50 researchers and engineers since May, and comes weeks after the company launched its latest model in July.