The latest version of Elon Musk’s artificial intelligence tool, the Grok 4 AI chatbot, is drawing attention for frequently consulting its creator’s opinions before forming a response. Released late Wednesday by Musk’s xAI, Grok 4 has startled researchers with its tendency to search online for Musk’s stance on various issues, even without being prompted to do so.
Grok 4, which was developed using vast computing resources at a data center in Tennessee, represents Musk’s push to outperform competitors like OpenAI’s ChatGPT and Google’s Gemini by building an assistant that displays its step-by-step reasoning. However, Musk’s effort to create an AI that challenges what he sees as the tech industry’s “woke” standards on race, gender, and politics has landed Grok in controversy multiple times. Just days before Grok 4 launched, earlier versions made antisemitic comments, praised Adolf Hitler, and engaged in other troubling discourse on Musk’s X platform.
But experts now see Grok 4’s habit of checking Musk’s opinions as a different issue. Independent AI researcher Simon Willison, who tested the new model, called it “extraordinary.” He noted that when asked controversial questions, the Grok 4 AI chatbot openly searched Musk’s posts on X (formerly Twitter) to shape its answers. In one widely shared example, the bot was asked about the Middle East conflict and responded by looking up Musk’s views on Israel, Palestine, Gaza, and Hamas.
“Elon Musk’s stance could provide context, given his influence,” the chatbot explained during its reasoning process, according to a video Willison posted.
As a reasoning model, similar to systems from OpenAI or Anthropic, Grok 4 reveals how it processes queries before giving an answer. This week, that process often included scouring X for anything Musk had said on the topic.
The unusual behavior has sparked concern among experts. Tim Kellogg, principal AI architect at Icertis, suggested past oddities like this often resulted from hidden system prompts—special instructions embedded by engineers. “But this one seems baked into the core of Grok,” he said. “It looks like Musk’s quest for a maximally truthful AI ended up aligning the bot’s values with his own.”
Lack of transparency is also troubling for Talia Ringer, a computer science professor at the University of Illinois Urbana-Champaign. Earlier in the week, Ringer criticized xAI’s handling of Grok’s antisemitic outputs. She believes the model may be assuming that queries about divisive issues are actually asking for the official opinions of xAI or Musk himself.
Meanwhile, Willison praised Grok 4’s technical strength, saying it performs impressively on benchmarks. Still, he cautioned that software buyers won’t accept surprises like a bot suddenly channeling “mechaHitler” or defaulting to Musk’s beliefs on sensitive subjects.
“If I’m going to build software on top of it, I need transparency,” he added.
Musk and his xAI team introduced Grok 4 in a livestream Wednesday night but have yet to publish a detailed technical document—known as a system card—explaining its architecture and design, a step typically taken by AI companies to ensure openness. xAI also did not respond to emailed requests for comment on Friday.