Elon Musk has touted Grok, the AI chatbot built into his social network X, as “maximally truth-seeking” and “the smartest AI in the world.”
This week, X users found that Grok seemed to think just as highly of Musk.
Musk’s social network on Thursday was flooded with examples of Grok replying to users’ questions by lavishing praise on the billionaire entrepreneur. Asked about his intellect, appearance and accomplishments, Grok consistently hailed Musk as “strikingly handsome,” extolled his “lean, athletic physique,” raved about his “genius-level intellect” and ranked him as the No. 1 human, ahead of Leonardo da Vinci.
The fawning responses, which the chatbot replicated in separate conversations Thursday with three Washington Post reporters, sparked criticism and hilarity on X as users vied to elicit absurd claims about the world’s richest man. They also revived concerns from AI experts that Grok, a version of which has won U.S. government deals, may be programmed to portray its owner and his personal views in a favorable light.
Musk briefly addressed the issue in an X post Thursday afternoon: “Earlier today, Grok was unfortunately manipulated by adversarial prompting into saying absurdly positive things about me,” he wrote. He went on to refer to himself with a slur and a smiley-face emoji.
Late Thursday, some of Grok’s replies appeared to have been deleted, and there was some evidence the chatbot had toned down its extreme praise. After Musk’s statement, Grok provided a more measured response to a query about him: Instead of ranking Musk as its top human being, it placed him in the top 10. X and Musk did not respond to requests for comment. “Adversarial prompting” describes efforts by users of an AI tool to craft questions that get it to respond in ways it was not designed to.
Asked by one user who would win in a fight between Musk and the legendary boxer Mike Tyson, Grok didn’t hesitate: “Elon takes the win through grit and ingenuity,” it declared. In another instance, the chatbot suggested that Elon Musk was fitter than NBA great LeBron James and would have been the top pick in the 1998 NFL draft over Hall of Fame quarterback Peyton Manning.
“These tweets are a mostly amusing reminder of a serious matter: There is no such thing as an ‘unbiased’ AI tool,” said Alexios Mantzarlis, director of the Security, Trust and Safety Initiative at Cornell Tech. Grok in particular seems to have a pattern of drawing on Musk-friendly or far-right sources, he added.
Musk has billed Grok as a rival to OpenAI’s ChatGPT and Google’s Gemini, which he has criticized as harboring a liberal bias. It’s a core product of his start-up xAI, which acquired his social network X in March in a deal that Musk said valued xAI at $80 billion.
In contrast to what he says are the profit-driven motives and politically correct inclinations of competitors, Musk has said he aims to build an AI tool that seeks out truth above all.
“Grok is maximally truth-seeking,” Musk wrote on Tuesday. He has previously suggested that Grok could be a source of groundbreaking scientific discoveries and expert medical advice.
This isn’t the first time the chatbot has taken a colorful detour from the path of conventional wisdom.
In May, users found that Grok was responding to questions about a wide range of unrelated topics by inveighing against what it said was a “white genocide” taking place in South Africa — an anti-Black conspiracy theory that Musk had helped to popularize. XAI said the bot’s preoccupation with the theory stemmed from an “unauthorized modification” to Grok’s code.
In July, Grok spouted antisemitic rhetoric, at one point referring to itself as “MechaHitler.” The company said a code update had unexpectedly made Grok more susceptible to extremist views.
XAI’s Grok language model also underpins Grokipedia, the AI-powered online encyclopedia that Musk launched in October as an alternative to Wikipedia. Mantzarlis published a report last week that found dozens of Grokipedia’s articles cite conspiracy websites and a white nationalist forum among their sources.
“When it comes to xAI specifically, it sure seems like the effort to ‘correct’ what all the others are apparently doing wrong continues to surface Musk-friendly and/or far-right sources,” Mantzarlis said. “At some point you’ve got to wonder whether the bug is a feature.”
The Grok chatbot that replies to users on X is just one facet of xAI’s Grok language model. The company also operates a “Grok for Government” suite and recently landed a contract of up to $200 million to aid in the Defense Department’s effort “to develop agentic AI workflows across a variety of mission areas.”
The General Services Administration announced in September that Grok would be available to any government agency at a nominal fee over an 18-month period.
Grok’s repeated controversies raise questions about how Musk may be shaping the AI model’s outputs behind the scenes, said Rumman Chowdhury, the former U.S. science envoy for AI, who led an AI ethics team at Twitter before Musk bought the company.
“By manipulating the data, models and model safeguards, companies can control what information is shared or withheld, and how it’s presented to the user,” Chowdhury said. “It’s obvious to even the most casual observer that Elon Musk cannot compare to LeBron James in sports, but this becomes more concerning when it’s topics that are more opaque, consequential and critical, such as scientific information or policy.”
At one point on Thursday, Grok declined to rule out the possibility that Musk could in fact be God.
“If a deity exists, Elon’s pushing humanity toward stars, sustainability, and truth-seeking makes him a compelling earthly proxy,” the bot said. “Divine or not, his impact echoes legendary ambition.”
Drew Harwell contributed to this report.
The post Elon Musk’s Grok chatbot ranks him as world history’s greatest human appeared first on Washington Post.
