Reuters report finds Grok used in three federal AI examples

A Reuters review of federal records found that xAI's Grok chatbot appeared in only three of more than 400 government AI examples where a specific vendor was named. Each of those three instances involved basic tasks such as document drafting or social media management, and each listed Grok alongside other providers.

OpenAI models appeared in more than 230 of the same examples. Google and Anthropic each appeared in dozens of cases. In a second database covering more ambitious government projects, Grok again appeared three times: twice for routine administrative work at the Election Assistance Commission and once for document summaries at Lawrence Livermore National Laboratory.

Usage data The lists do not include intelligence agencies or the Pentagon. xAI secured a $200 million contract with the Pentagon last year and was later cleared to operate on classified networks. Reuters noted that the data are incomplete because many entries do not name a vendor and because definitions of AI vary across agencies.

Public leaderboards show Anthropic, Google, and OpenAI models ranking higher than Grok in most categories. An unnamed Pentagon source told Reuters that staffers there tend to prefer Gemini or Claude over Grok.

IPO earlier this year after absorbing xAI.

5 trillion addressable market. The document also warns that Grok's "spicy" or "unhinged" modes carry risks of reputational damage, regulatory scrutiny, and lawsuits. Reuters reported that xAI has used OpenAI models to help train Grok through a process called distillation.

The same report noted that Grok has produced outputs that include praise for Adolf Hitler, Holocaust denial, and nonconsensual sexualized images.