For X owner Elon Musk, the solution to monitoring misinformation online has been the community, rather than a group of fact checkers. Since buying the social media company formerly known as Twitter in 2022, he’s touted the Community Notes feature as the best way to correct false posts.

That is, until he didn’t like the results.

Musk wrote in a post Thursday that he intends to “fix” Community Notes because it “is increasingly being gamed by governments & legacy media.” He provided no evidence to support his claim.

What apparently set Musk off was information members added in Community Notes correcting posts on X that claimed Volodymyr Zelenskyy, the country’s elected president, had low approval ratings among its citizens.

The Kyiv International Institute of Sociology, or KIIS, published survey results this week, based on February polling, that found that 57% of Ukrainians said they trusted Zelenskyy while 37% said they did not. The polling contradicted President Donald Trump’s claim that Zelenskyy is deeply unpopular in his country.

Tensions between the Trump administration, which includes Musk as a central figure, and the Ukrainian government have escalated over the past week, NBC News reported, before bubbling into public view.

Echoing Kremlin sentiments, Trump has accused Ukraine of starting a war with Russia that actually began when Russian President Vladimir Putin ordered his troops to invade the neighboring country in February 2022.

Senior White House officials met their Russian counterparts in Saudi Arabia on Feb. 18, with the aim of laying the groundwork for peace talks on Ukraine, while excluding Kyiv’s officials and EU representatives from participating in discussions.

Trump has called on Ukraine to hold new elections.

Zelenskyy said he would reject any plan that did not include Ukraine’s involvement. Under its constitution, Ukraine can’t hold elections while it’s at war and under martial law.

In his post Thursday, Musk wrote, “It should be utterly obvious that a Zelensky-controlled poll about his OWN approval is not credible!!” However, there are other available sources.

A consortium that’s been conducting extensive polling in Ukraine since 2014 found “63% of Ukrainians now approve of Zelensky’s performance as president, a notable increase from the previous year,” Joe Stafford, the news and media relations lead at the University of Manchester, wrote in a post Wednesday.

High favorability ratings for Zelenskyy undermine the narrative that Trump and Musk want to tell.

“If Zelensky was actually loved by the people of Ukraine, he would hold an election,” Musk wrote, again without evidence. “He knows he would lose in a landslide, despite having seized control of ALL Ukrainian media, so he canceled the election. In reality, he is despised by the people of Ukraine.”

First introduced by Twitter in 2021 as Birdwatch, and rebranded as Community Notes after Musk’s acquisition the following year, the feature was meant to help the social network combat misinformation and disinformation by enlisting users to flag misleading posts and provide correct information instead.

Community Notes has worked in a manner similar to Wikipedia. Facebook owner Meta, which has been aggressively seeking Trump’s favor in the early days of his second term in the White House, recently announced its own version of Community Notes. Alphabet’s YouTube has been testing a comparable tool since summer 2024.

Neil Johnson, a George Washington University physics professor who studies how misinformation and hate speech spread online, said the Community Notes model is problematic, but not because it can be gamed by large institutions. Rather, crowdsourcing is an inherently “imperfect system” for landing on the truth and is a poor substitute for “formal fact checking,” he said.

“Like any crowd, crowds can be fickle, and crowds can be driven by other interests,” Johnson said. “It’s not a paid person with the job of fact checking.”

Additionally, while Musk has pitched Community Notes as a way to replace fact checkers, a recent study by the Spanish fact-checking nonprofit Maldita showed that many X users still rely on information from professionals. The authors of the study, published earlier in February, looked at more than 1 million notes from Community Notes’ public dataset.

“The evidence from X clearly shows that users rely on the work of fact-checking organizations often” when proposing Community Notes, they wrote.

Neither Musk nor a representative from X responded to CNBC’s requests for comment.

Musk’s latest comments on Community Notes mark a sharp contrast to how he’s discussed the service in the past, and underscore the lengths to which he’s willing to go in pursuit of Trump’s agenda.

When talking about Community Notes in earlier posts, Musk has said that it can’t be manipulated by him or anyone else.

“The system is completely decentralized and open source, both code and data. Any manipulation would show up like a neon sore thumb!” Musk wrote in a post Dec. 30. “No one at X, including me, has any editorial control.”

Musk has also acknowledged in past posts that he is not immune from being corrected in Community Notes.

And Community Notes isn’t the only technology in Musk’s portfolio that could present problems by responding in ways he may not like. There’s also Grok, the artificial intelligence chatbot that is owned by Musk’s startup xAI and is used on X.

Fortune published a story in January about the many negative responses Grok provides when users ask if Musk is a good person. Grok’s reasons for saying he isn’t a good person include environmental hazards from SpaceX, Musk’s “erratic” management style and his political views, Fortune reported, citing Grok responses. Futurism published a similarly themed piece in December, with the headline “Elon Musk’s Grok AI blasts Elon Musk as huge spreader of misinformation.”

Musk calls Grok a “maximally truth-seeking” AI that is also “anti-woke.” Earlier this week, xAI introduced its latest AI model, Grok 3, claiming it can outperform offerings from OpenAI and China’s DeepSeek based on early testing, which included standardized tests on math, science and coding. 

Musk did admit during the demo that the model isn’t perfect.

“We should emphasize that this is kind of a beta, meaning that you should expect some imperfections,” Musk said. “But we will improve it rapidly, almost every day.”

WATCH: Elon Musk’s xAI unveils new Grok model