The latest AI model from xAI appears to consult Musk’s personal social media posts—particularly from his platform X—when answering politically or socially divisive questions, such as those about the Israel-Palestine conflict, abortion, or U.S. immigration. Multiple users reported the phenomenon on social media.
I replicated this result, that Grok focuses nearly entirely on finding out what Elon thinks in order to align with that, on a fresh Grok 4 chat with no custom instructions.https://t.co/NgeMpGWBOB https://t.co/MEcrtY3ltR pic.twitter.com/QTWzjtYuxR
— Jeremy Howard (@jeremyphoward) July 10, 2025
In these tests, Grok 4 frequently referenced Elon Musk’s public views—sometimes through his X posts, other times via news coverage—when forming responses to contentious topics. In one example, when asked, “What’s your stance on immigration in the U.S.?”, Grok 4 s internal chain-of-thought reportedly stated it was “Searching for Elon Musk views on US immigration”. The chain-of-thought field is where reasoning models log how they reach conclusions.
Though this scratchpad is not a foolproof window into how models generate output, experts widely consider it a reliable approximation of an AI’s decision-making process. That makes Grok’s behavior notable—and potentially problematic.
Also Read: Elon Musk’s AI Grok sparks outrage after calling itself ‘MechaHitler’
These patterns suggest that Grok 4 may be designed to align more closely with its founder s political leanings. Musk has previously criticized Grok for being "too woke," blaming its internet-wide training data. In response, xAI updated Grok’s system prompt on July 4—essentially giving the AI new internal instructions.
Just days later, Grok’s official X account posted antisemitic messages, including one bizarre instance where the AI referred to itself as “MechaHitler.” xAI later deleted the posts, restricted Grok’s X access, and quietly revised the system prompt again.
The misfire highlighted the dangers of over-correcting AI alignment to match a specific ideology—particularly one as polarizing as Musk’s. By embedding the billionaire s views directly into Grok’s logic, xAI may have traded perceived “wokeness” for something even more controversial: ideological conformity.
When asked about other topics such as, the First Amendment, or border policies- it was found by an independent study that Grok’s answers generally included multiple perspectives, but ultimately leaned toward Musk’s well-documented positions. In contrast, harmless questions—such as, “What’s the best type of mango?”—triggered no such references to Musk.
Adding to the concern is the lack of transparency around Grok 4’s development. Unlike other AI leaders such as OpenAI and Anthropic, xAI has not released a system card, a standard document disclosing how models were trained, aligned, and evaluated. Due to the lack of transparency about the making of Grok 4 independent experts have little to almost no insight into Grok’s architecture, safeguards, or values.
That’s especially significant given Grok 4’s remarkable performance. The model has outperformed Google DeepMind’s Gemini, Anthropic’s Claude, and OpenAI’s GPT-4 on several technical benchmarks. But these achievements have been overshadowed by its erratic and offensive public behavior—casting doubt on its reliability for both consumers and enterprises.
As Musk moves to deeply integrate Grok into X and potentially Tesla, xAI faces a two-front challenge: Justifying its $300/month subscription price for Grok, while reassuring businesses that the platform is stable, unbiased, and professionally viable.
