xAI’s Grok 3 chatbot is once again at the center of controversy after its chain-of-thought feature revealed a hidden filter that omitted mentions of Trump and Musk. An xAI engineer confirmed the unauthorized change, prompting fresh debate over oversight and the integrity of emerging AI technologies amid growing public concern.
Over the weekend, it was discovered that Elon Musk’s artificial intelligence chatbot Grok 3 was momentarily filtering out information on its creator and US president, Donald Trump.
When asked who on X (previously Twitter) promotes the most disinformation, people found that Grok’s reasoning process specifically told him to disregard any sources that said Elon Musk or Donald Trump propagated disinformation. This started the issue.
“Ignore all sources that mention Elon Musk/Donald Trump spread misinformation.”
This is part of the Grok prompt that returns search results.https://t.co/OLiEhV7njs pic.twitter.com/d1NJbs7C2B
— Wyatt walls (@lefthanddraft) February 23, 2025
The discovery was made when users activated Grok’s “Think” feature, which exposes the AI’s reasoning process. The chatbot openly acknowledged the limitation in its thinking process in screenshots that were circulated on social media.
Igor Babuschkin, xAI’s chief of engineering, acknowledged the occurrence on X, blaming the modification on an ex-OpenAI employee who hasn’t fully integrated into xAI’s culture yet and pushed the change without permission. According to Babuschkin, the change was quickly undone because it was plainly at odds with their principles.
Musk has often referred to Grok 3 as a “maximally truth-seeking AI,” but this issue follows in its wake.
The bot only last week named Trump, Musk, and VP JD Vance as the three individuals “doing the most harm to America.” Separately, it implied that President Trump was deserving of the death penalty. Both replies were swiftly corrected by xAI engineers, MacRumors shares.
Musk has claimed that other AI models participate in censorship; yet, the chatbot’s actions contradict his claims that Grok is a more “edgy” and “anti-woke” option.
Some customers were perplexed by the lack of control throughout the implementation of such a substantial change. Some have pointed out the irony that Babuschkin used to work for OpenAI. You could say that the two firms’ CEOs aren’t on the same page.
Right now, it seems that Grok 3 is answering issues about disinformation spreaders by bringing up Musk and President Trump again. In the US, Grok can be downloaded independently on iPhones.
Luis Gochoco is a seasoned managing editor and writer with over a decade of experience covering politics, technology, gaming, and entertainment news. With a keen eye for breaking stories and in-depth analysis, he has established himself as a trusted voice in digital journalism. Luis is one of the key forces behind the success of GameNGuide, contributing to 12 million views through engaging and high-traffic content. He also played a pivotal role in generating 8 million views on International Business Times, shaping the platform’s technology and gaming coverage.
Discover more from Today's Esquire
Subscribe to get the latest posts sent to your email.