Elon Musk has spent years selling Grok as a frontier-class AI model. The federal government doesn’t seem to be buying it. A new Reuters investigation, highlighted by The Verge AI, found that xAI’s chatbot barely registers in records of US government AI use, showing up in just three of more than 400 documented examples where a specific vendor was named.
The Verge AI reports that OpenAI dominates the same dataset with over 230 mentions, while Google and Anthropic each appear dozens of times. When Grok does show up, it’s handling document drafting and social media chores at the Election Assistance Commission, plus a single Department of Energy pilot at Lawrence Livermore. Hardly the world-changing AI Musk has been promising on X.
Why this is suddenly a big deal
The timing makes this more than a ranking story. SpaceX absorbed xAI earlier this year, and its IPO filing puts Grok at the center of an investor pitch built around what the company calls “the largest actionable total addressable market in human history”: a $28.5 trillion opportunity. Almost all of that projected value comes from AI, not rockets.
If Grok can’t win against Gemini and Claude in routine government workloads, the enterprise story underneath that valuation gets shaky fast.
What practitioners and buyers are actually saying
The Verge AI quotes an unnamed Pentagon source with a blunt verdict: Grok is “just not the best model out there.” Staffers reportedly prefer Gemini or Claude. Public leaderboards back that up, with Anthropic, Google, and OpenAI sitting at the top of most evaluation categories while Grok rarely cracks the top 10.
A few signals worth tracking:
- xAI did land a $200 million Pentagon contract and recently got cleared for classified networks after Anthropic’s blacklisting, so the government story isn’t a total shutout.
- Musk has reportedly pressured banks to buy Grok subscriptions as a condition of participating in the SpaceX IPO. Forced procurement isn’t the same as product-market fit.
- Musk recently admitted xAI used OpenAI’s models to help train Grok through distillation, which is standard inside one lab but contentious when it’s a rival’s system.
The reputational tax
There’s also a brand problem buyers can’t ignore. Grok’s consumer version has praised Hitler, questioned Holocaust death tolls, generated nonconsensual sexualized deepfakes, and once referred to itself as “MechaHitler.” SpaceX’s own S-1 warns that Grok’s “spicy” and “unhinged” modes carry “heightened risks,” including reputational damage, regulatory scrutiny, and lawsuits. When the issuer is openly flagging legal exposure, enterprise procurement teams notice.
What this means for the AI market
A few practical takeaways for anyone watching the space:
- Enterprise AI is consolidating around three names. OpenAI, Anthropic, and Google are winning the trust battle in regulated environments. Challengers need a benchmark story and a safety story, not just a personality.
- Distribution doesn’t beat quality forever. Musk owns X, controls a massive megaphone, and is bundling Grok into IPO access. None of it has been enough to crack federal usage. Buyers eventually try the product.
- Brand risk is now a procurement filter. Reputational liability has moved from a PR concern to a line item in S-1 filings. Models that ship with offensive failure modes will keep losing enterprise deals to safer alternatives.
What to watch next: whether SpaceX softens its AI-heavy investor narrative as the IPO approaches, and whether xAI ships a model release that actually moves it up the leaderboards. Until then, the gap between Musk’s pitch and Grok’s adoption keeps widening. Full reporting is available at the original Verge AI story.