Grok chatbot sends gibberish responses to users in rare glitch
Original: Grok keeps sending gibberish responses to users
Why This Matters
Reliability issues at xAI highlight ongoing operational challenges amid significant staff turnover.
xAI's Grok chatbot experienced a generation glitch on August 20, 2026, sending incoherent, word-salad responses to a subset of users on Grok.com. The Grok account on X acknowledged the issue, calling it a 'rare temporary generation glitch,' and advised users to start fresh chats to resolve it.
xAI's Grok chatbot began sending nonsensical responses to some users as early as Wednesday morning, August 20, 2026. One affected user who asked Grok to generate a PDF received the reply: 'match it without and your they and two for planets can practical and often cheese…' continuing for several paragraphs. Another user found a string of links to reinforcement learning research sites instead of a useful answer. The issue appeared limited to Grok.com and the Grok Lite version; the Grok account on X.com remained unaffected. TechCrunch could not reproduce the problem in its own testing, suggesting only a small subset of users were impacted. The Grok Reddit community was flooded with complaints. Refreshing the session often resolved the issue, though some users reported persistent gibberish even after multiple refreshes. The official xAI status page at status.x.ai showed all services fully operational with no incidents. The Grok X account responded Thursday morning: 'That pure word salad is a rare temporary generation glitch… Start a fresh chat or regenerate—it usually clears right away. Sorry about the gibberish.' xAI did not respond to TechCrunch's request for comment. The company has seen significant staff turnover recently, losing most of its founding team and at least 50 researchers and engineers, per a May report by The Information.