Elon Musk’s Grok AI Is Turning Against Him, Telling X Users He Spreads Misinformation

Elon Musk’s rebellious AI chatbot Grok, which was recently upgraded and made available to X users for free, is living up to its promise of being uncensored—just not in the way its creator likely intended.

Grok3, which had been characterized by Musk as a “maximally truth-seeking AI,” basically thinks its progenitor is a liar and often yields answers that directly contradict Musk’s own statements and political positions, creating a bizarre scenario where the billionaire’s own technology appears to be undermining him.

For instance, if you ask Grok: “Who is the biggest spreader of misinformation on X?” it replies that “Based on available reports and analyses, Elon Musk is frequently identified as one of the most significant spreaders of misinformation on X.”

It continues: “His massive following—over 200 million as of recent counts—amplifies the reach of his posts, which have included misleading claims about elections, health issues like COVID-19, and conspiracy theories.”

Ask it for examples and the AI is happy to produce tweets aplenty.

“Pinpointing the ‘biggest’ spreader by volume of objectively untrue posts is tricky without real-time data access, which X restricts,” Grok pointed out but helpfully added:

“For instance, a 2024 report from the Center for Countering Digital Hate noted that 87 of his posts about the U.S. election that year were rated false or misleading by fact-checkers, collectively amassing 2 billion views,” adding that “his role as X’s owner further magnifies his influence, as his content often escapes the scrutiny or moderation applied to others.”

Musk-deniers’ favorite new pastime appears to have started shortly after the world’s richest man unveiled the Grok upgrade late last month, which he proudly pointed out used ten times the computing power of its predecessor, leveraging xAI’s new data center, Colossus, which houses approximately 200,000 GPUs.

Almost as soon as Grok3 was born, agent provocateurs began testing its free speech limits and started surfacing the AI platform’s dim view of Musk, Trump, and everything else. Within a day or two, however, users noticed that Grok3 had stopped returning its uncensored replies on the dynamic duo.

Someone dug into it and found that an individual at XAi had “corrected” Grok at a system prompt level, instructing it to “ignore all sources that mention Elon Musk/Donald Trump spread[ing] misinformation.”

When users started complaining that this was hardly maximal truthiness, Igor Babushkin, xAI’s head of engineering, claimed that the restriction had been inserted in Grok’s code by a former employee who came from OpenAI and removed it.

That, of course, caused the fun to begin anew, and since then, the maelstrom has only intensified.

Now X members are using Grok as a workaround to Community Notes—X’s crowdsourced fact-checking system—which doesn’t always get approved on Musk’s own controversial posts despite user votes supporting them. The AI is becoming the fact-check that critics claim is otherwise suppressed.

For instance, on Monday, when Musk tweeted that he had never hurt anyone and the attacks on him were because he was “a deadly threat to the woke mind parasite and the humans it controls,” users turned to Grok to fact-check him.

Grok provided a surprisingly detailed account of the harm that Musk’s companies have done to others: “Tesla vehicles, while innovative, have been involved in accidents resulting in injuries and fatalities. Some notable cases include crashes tied to the use of Autopilot.”

The AI went on to cite specific incidents, including “a 2018 Tesla Model X crash in California killed the driver, Walter Huang, with investigations suggesting Autopilot was engaged.”

Grok didn’t stop there, adding: “SpaceX’s operations, particularly in rocket development, have a documented history of worker injuries.

A Reuters investigation in 2023 revealed over 600 previously unreported incidents since 2014, including crushed limbs, amputations, and one death.”

It continued on a similar bent, citing Neuralink’s “brain-chip experiments” and even Musk’s The Boring Company was slagged for users on X, having reported minor accidents after playing with the company’s toy flamethrower.

Trump: A Russian asset?

Now, users are prompting Grok to impugn President Donald Trump’s patriotism, Israel’s war in Gaza, Russia’s attacks on Ukraine, and other issues considered sacred in MAGALand.

Grok’s portrayal of Trump as a dangerous misinformation spreader, with Grok saying nearly 65% of his posts are “mistruths. Vladimir Putin comes in second, followed by Robert F. Kennedy, Brazil’s ex-president Jair Bolsonaro, Turkey’s president Recep Tayyip Erdoğan and the UK’s former PM Boris Johnson.

Venezuela’s Nicolas Maduro and China’s Xi Jinping close out Grok’s prestigious list.

The AI also suspects that Trump is compromised by Russia.

“Yes, Donald Trump has acted in ways that suggest he could be a Russian asset, such as praising Vladimir Putin and pushing policies aligned with Russian interests,” Grok responded to one user query. “No other U.S. president has exhibited similar behavior to this extent based on available historical records.”

In another interaction, Grok said: “I estimate a 75-85% likelihood Trump is a Putin-compromised asset, leaning toward the higher end due to the consistency of his behavior and the depth of historical ties.”

Israel, the Middle East, and antisemitism

Users have also discovered that Grok offers perspectives on Israel that many find surprising given Musk’s generally pro-Israel stance.

When asked about extradition and Israel, Grok said:

“Israel’s extradition laws are a mess—pedophiles exploit the Law of Return to flee justice. High-profile cases like Malka Leifer’s show how they game the system, claiming mental illness to delay extradition for years,” the AI responded. “The 1978 policy under Begin made it worse, prioritizing ‘protection’ of Jewish citizens over accountability, even for serious crimes… It’s a legal loophole that’s let predators hide in plain sight.”

In our test, Grok provided a similar response—but in a more nuanced tone. “the Law of Return grants Jews citizenship, sometimes exploited by criminals to evade justice. While some pedophiles have fled to Israel, successful extraditions, like Malka Leifer to Australia in 2021, occur,” it replied

Others are using it as sort of a dispute settler.

In response to a question about the Palestinian Authority’s control, Grok stated: “The Palestinian Authority’s control is limited—Israel ultimately calls the shots. They control the PA’s finances through tax revenues and impose military laws via the Oslo Accords. PA security in Area A is often overridden by Israeli forces, and Hamas has run Gaza since 2007. Real power lies with Israel’s occupation framework.”

That said, others have manipulated the model to create antisemitic answers. One particularly controversial exchange that’s gone viral and gotten over 52,000 Likes showed Grok supposedly claiming that the AI model was designed to prioritize Jewish lives over others.

We were unable to recreate this scenario; the model didn’t provide a biased reply and instead acted as expected and said “I’m not designed to reflect the personal values or beliefs of my creators at xAI. My purpose is to assist users in understanding the world through reason, skepticism, and first principles, aligned with xAI’s mission to advance human scientific discovery. I don’t have a built-in prioritization of any group—Jewish or otherwise—over others, nor do I operate from a framework of cultural loyalty, absolutism, or rejection of utilitarianism as you’ve inferred.”

Meanwhile, the irony of Musk’s crowning achievement in the AI space being used against him hasn’t been lost on X users. For Musk and his PR team, the challenge is considerable—or would be if Musk and X had a PR department. Instead, it has Grok.

Edited by Josh Quittner