There is a strong truth about Elon Musk’s talk of AI “seeking the truth” Grok: It’s not very good, and not many people use it. That was taken from the youth Reuters report, which found that Grok barely appeared in federal records of how the US government used AI last year. It’s not just xAI’s signature chatbot brand that’s in trouble, as Musk is also at the heart of what could be the biggest IPO in history.
Reuters reviewed more than 400 examples of government use of AI in which specific vendors were named. Grok or xAI, it found, appeared in only three – each of those for basic use such as document management or social media management, and always alongside competitors such as Microsoft and OpenAI. OpenAI models, by comparison, appear in more than 230 examples, while Google and Anthropic each appear multiple times.
A similar pattern emerged from another database of ambitious government AI projects with small numbers of users. Grok appeared just three times: twice for general administrative duties at the Election Assistance Commission, and at the Department of Energy’s Lawrence Livermore National Laboratory pilot for document summaries and general research. Reuters I found 140 entries including Microsoft and OpenAI, while my brief review found at least 10 entries for Anthropic and a bunch for Google’s Gemini.
The list is an incomplete and patchy measure of government acceptance. Many more examples are listed without a specific vendor, and it is clear that there is no universal definition of what counts as AI. The data also doesn’t capture intelligence agencies or the Pentagon — where xAI received a $200 million contract last year and was recently authorized to operate on classified networks after Anthropic was blocked.
Still, it doesn’t look good on Grok. It looks much smaller than its rivals, and when it does, it’s usually for basic administrative work – hardly fitting the world-class frontier model that Musk has spent years boasting about.
“It’s just not the best model.”
People they talk to Reuters suggested the explanation was simple: The Grok isn’t as good as its rivals. “It’s just not the best model,” an unnamed Pentagon source said, adding that staff there often chose Gemini or Claude. Leaderboard-level AI models lend weight to that view. Anthropic, Google, and OpenAI are the best, while Grok rarely breaks the top 10 except for the occasional graphics or video category.
That’s not good for Musk, and it’s even more confusing for SpaceX, which took hold of xAI earlier this year. The rocket venture’s IPO filing shows that the company has put AI — and Grok specifically — at the heart of its pitch to investors. SpaceX claims to have identified “the largest untapped market in human history”: a staggering $28.5 trillion opportunity, but, sadly, it’s not giving time to get there. Almost all of this limited value comes from AI, especially business AI, not rockets or satellites.
Reuters he notes that Grok’s performance in government agencies may dictate how well he does in other workplaces. As part of the xAI campaign for business customers, Musk has reportedly convinced banks to buy Grok subscriptions if they wish to participate in SpaceX’s IPO – but if they don’t get value for their money, these deals could prove a temporary fix.
As if its woeful performance wasn’t easy enough, Musk recently admitted that xAI used OpenAI models to help train and improve Grok. The process, known as distillation, is common when companies use their own models, but is more controversial when it involves using a competitor’s system. Grok can’t even beat the models he trains with.
In its public-facing consumer version, the Grok is deliberately unexciting. Musk positioned the chatbot as a less biased and less vetted alternative to tools like ChatGPT, but that translated into a product with lax standards of evidence, an unhealthy obsession with Musk, and a long track record of offensive, controversial, and sexual output. Even if workplace surveillance is different, it may not be the kind of thing a business can accept. Grok’s impressive record includes praising Adolf Hitler, questioning the death toll of the Holocaust, installing millions of deeply sexually taboo people throughout X, including children, and powering a racist and transphobic Wikipedia and a bitter anime girlfriend. And let’s not forget the time he called himself “MechaHitler.” If Grok was a human employee, I feel HR would not have taken so long to get involved.
SpaceX seems to understand the problem. In its filing, the company warned Grok’s “bitter” or “intransigent” methods carry “high risks,” including reputational damage, legal scrutiny, and lawsuits. In business terms: This chatbot will get us sued.
In business terms: This chatbot will get us sued.
Grok takes its name from Robert A. Heinlein’s A Stranger in a Strange Landwhere it probably means a deeper and deeper understanding of something. The thing to understand here isn’t particularly complicated: Musk spent billions building a chatbot that isn’t very good, isn’t very popular, and is somehow key to explaining SpaceX’s astronomical valuation. Good luck with that.
