Grok has started giving strange, nonsensical replies to some users this week. The chatbot, made by Elon Musk's company xAI, appears to be generating word salad instead of normal answers.
One user asked Grok to create a PDF. Instead they got a reply that read like scrambled words with no clear meaning.
Another user checked the source links attached to their response. They found a string of links pointing to reinforcement learning research sites instead of anything related to their question.
People who spoke with TechCrunch said they were using Grok Lite when they saw the issue. Several said it started as early as Wednesday morning.
TechCrunch tried to recreate the problem during its own testing but could not. This suggests the bug may only affect a small group of users.
xAI did not respond to a request for comment on the issue.
What Users Are Reporting
Grok's community on Reddit has filled up with complaints about the glitch. Many users shared screenshots of confusing or broken responses.
Some people said refreshing their chat session fixed the problem right away. A new session often returned normal answers.
Other users said the gibberish kept happening even after they refreshed several times. For them, the fix did not work consistently.
So far, the bug only seems to affect direct queries typed into Grok.com. The separate Grok account built into X has not shown the same issue.
xAI's Response So Far
The Grok account on X addressed the complaints directly on Thursday morning. It described the issue as a rare and temporary generation glitch.
The account pointed users to xAI's status page. That page showed all Grok services listed as fully operational with no incidents reported.
The post suggested starting a fresh chat or asking Grok to regenerate its answer. It said this usually clears the problem right away.
The account also apologized to users for the confusing responses. It did not give a technical explanation for what caused the glitch.
This bug arrives during a period of change at xAI. The company has seen a wave of staff departures over the past several months.
A report from The Information in May said xAI lost most of its founding team. The same report said at least 50 researchers and engineers had also left the company.
Despite the turnover, xAI has kept releasing new models. In July, the company put out its newest foundation model.
xAI described that July release as an Opus-class model. The company said it was faster, more token efficient, and lower cost than earlier versions.
As of Thursday, xAI had not issued a separate statement explaining the root cause of the gibberish responses. The X account's message remains the only official comment on the issue so far.
