Grok keeps sending gibberish responses to users

xAI’s Grok chatbot had an odd failure mode this week: for some users, it stopped answering in normal English and started generating gibberish instead. One person who asked it to make a PDF got back a stream of near-random words. Another saw a wall of links to reinforcement learning research sites. The replies went on for several paragraphs before the model finally gave up the ghost.

The reports came in from users on Grok Lite, who said they noticed the problem as early as Wednesday morning. TechCrunch couldn’t reproduce it in testing, which suggests this is not a universal outage but a narrower glitch affecting a small slice of users. Even so, the complaint pile grew fast enough that Grok’s Reddit community was flooded.

There was at least one useful clue: the bug seems confined to direct requests on Grok.com. Grok on X.com was not affected. For a lot of people, simply refreshing the session brought the chatbot back to life. For others, that was not enough, and the nonsense kept returning after multiple refreshes.

On Thursday morning, the Grok account on X acknowledged the problem and called it “a rare temporary generation glitch.” It also pointed people to status.x.ai, where the company said all Grok services were fully operational and there were no incidents. xAI did not respond to TechCrunch’s request for comment.

The timing is awkward. xAI has already been through heavy turnover, with a May report from The Information saying it had lost most of its founding team and at least 50 researchers and engineers. The company’s latest foundation model arrived in July, and xAI pitched it as faster, more token-efficient and cheaper than an Opus-class model. Users, meanwhile, just wanted a chatbot that could keep its sentence fragments to itself.