What happened in the August 20, 2026 incident in which xAI’s Grok chatbot sent gibberish “word salad” responses—particularly to users of Grok Lite who asked it to generate PDFs—including what the nonsensical output and source-link behavior looked like, when and where the problem

On August 20, reports described an intermittent Grok.com generation failure affecting some Grok Lite users, especially after PDF-generation requests. Instead of producing a PDF or coherent answer, the chatbot could emit several paragraphs of grammatically broken, unrelated English words; xAI did not publicly provide a technical root cause. 8

• What users saw: One reported output began, “match it without and your they and two for planets can practical and often cheese…,” then continued as multi-paragraph word salad. In at least one case, the response’s source citations led to a string of reinforcement-learning research sites rather than meaningful supporting sources. 8

• Timing and location: Users said they had seen it from Wednesday morning, August 19; the story was published August 20. Evidence pointed to direct prompts on the Grok.com web service, not every Grok endpoint. 8

• Scope and reproducibility: Complaints were numerous enough to swamp Grok’s Reddit community, and affected people who spoke to the press said they were using Grok Lite. But the issue was not reliably reproducible: TechCrunch could not trigger it in its own tests and assessed it as likely limited to a small subset of users. Some people recovered after refreshing, while others said repeated refreshes did not help. 8

• Suggested remedies: Users found that refreshing the session often restored normal output. Grok’s official X account called it a “rare temporary generation glitch” and advised starting a fresh chat or regenerating the answer, saying that usually cleared it. 8

• What appeared unaffected: Grok’s account/functionality on X.com appeared unaffected, reinforcing the view that this was a Grok.com-specific failure. 8

• Official signals: Grok’s X account said xAI’s status page showed all services operational and no declared incident. Meanwhile, xAI itself did not respond to TechCrunch’s request for comment, so there was no independently confirmed explanation, severity assessment, or postmortem. 8

• Broader context—but not proven causation: The incident occurred after substantial reported turnover: xAI had lost most of its founding team and at least 50 researchers and engineers, according to reporting cited by TechCrunch. It also followed the July release of Grok 4.5, which the company characterized as its most capable, “Opus-class” foundation model for coding and agentic work. Those facts provide operational context, but there is no evidence that staff departures or the Grok 4.5 release caused this specific glitch. 82