An unusual Grok bug has left some users staring at responses that look less like answers and more like scrambled fragments of language. The issue has affected people using Grok Lite, according to users who spoke to TechCrunch, though TechCrunch said it could not reproduce the problem in its own testing.
The reports point to a limited but visible failure in the chatbot experience: direct questions on Grok.com sometimes returned gibberish instead of useful replies. The Grok account on X.com, however, has not been affected.
What Users Saw
One example came after a user asked the model to generate a PDF. Instead of producing a useful result, the chatbot responded with a nonsensical string that began: match it without and your they and two for planets can practical and often cheese... Similar text reportedly continued for several paragraphs.
Another user encountered a different kind of odd output while checking source links. In that case, the response contained a string of links to reinforcement learning research sites.
Affected users told TechCrunch they were using Grok Lite. Some noticed the problem as early as Wednesday morning. TechCrunch described the issue as likely affecting only a small subset of users, since its own testing did not recreate the bug.
Where The Problem Appears Limited
The reports suggest the glitch is not hitting every part of Grok. The issue appears to be limited to direct queries on Grok.com. The Grok account on X.com has continued operating without the same problem.
That distinction matters because it narrows the apparent surface of the bug. Users seeing gibberish in one place may still find that other Grok-related experiences are not behaving the same way.
The Reddit community for Grok has filled with complaints from confused users. For some, refreshing the session brought the chatbot back to normal. Others said the nonsense responses kept appearing even after multiple refreshes.
What Grok Said About The Glitch
The Grok account addressed unhappy users on X and confirmed that something had gone wrong. It described the behavior as a rare temporary generation glitch while also pointing users to the official status page.
That pure word salad is a rare temporary generation glitch. Official status at https://status.x.ai shows all Grok services fully operational with no incidents. Start a fresh chat or regenerate—it usually clears right away. Sorry about the gibberish,
The message gives users two immediate actions: start a fresh chat or regenerate the response. The account said that usually clears the issue right away.
At the same time, the status page reference shows a mismatch between user-visible failure and official service status. According to the Grok account, all Grok services were fully operational with no incidents, even while some users were receiving incoherent answers.
Why This Kind Of Failure Gets Attention
For users, the problem is simple: a chatbot that returns gibberish breaks the basic promise of the product. Whether the request is to generate a PDF, check links, or answer a direct query, the user expects coherent output that can be evaluated and acted on.
The issue also stands out because it is not merely a wrong answer. A wrong answer can be hard to spot. A response made of disconnected words is immediately visible, confusing, and difficult to recover from if refreshes do not solve it.
That is why even a problem affecting a small subset of users can become highly visible. When screenshots or examples circulate in a community, the failure becomes part of the broader conversation about whether the tool is reliable in ordinary use.
The Broader Context Around X.ai
The glitch comes after a period of significant staff turnover at X.ai. According to a report from The Information in May, the company has lost most of its founding team and at least 50 researchers and engineers.
X.ai also released its most recent foundation model in July. The company described that model as an Opus-class model, but faster, more token-efficient and lower cost.
TechCrunch said X.ai did not respond to a request for comment. For now, the clearest public guidance remains the Grok account's advice: begin a new chat or regenerate when the chatbot produces word salad.
The practical takeaway is narrow but important. The reported bug appears limited, but users who rely on Grok Lite or direct queries on Grok.com may need to watch for malformed output and retry when the chatbot's response stops making sense.