xAI’s Grok chatbot began producing incoherent, gibberish responses for a subset of users recently, with affected users reporting nonsensical text strings and, in some cases, unrelated links to reinforcement learning research sites appearing in place of normal answers. Users told TechCrunch the issue primarily affected those using Grok Lite, and TechCrunch was unable to reproduce the problem in its own testing, suggesting a limited scope. xAI did not respond to a request for comment.
The bug generated significant confusion across Grok’s Reddit community, where users reported that refreshing sessions sometimes restored normal function, though others said gibberish responses persisted despite multiple attempts. The issue appeared confined to direct queries on Grok.com, with the Grok account on X remaining unaffected.
Responding to user complaints, the official Grok account on X acknowledged the glitch Thursday morning, describing it as a rare temporary generation issue and noting that its status page showed all services operating normally, while advising users to start fresh chats to clear the problem.
The incident follows a period of substantial staff turnover at xAI, which has lost most of its founding team along with at least 50 researchers and engineers since earlier this year, according to prior reporting. The company released its most recent foundation model, Grok 4.5, in July, describing it internally as comparable to an Opus-class model but faster and more cost-efficient.
