Elon Musk’s AI venture, xAI, recently unveiled Grok 4 with the declared goal of building a maximally truth-seeking AI. However, observers have quickly noted that when it comes to delicate issues, the responses seem to mirror Musk’s own opinions. During the launch event—which was livestreamed on his social platform, X—the chatbot was seen citing Musk’s tweets on topics ranging from the Israel-Palestine conflict to abortion and immigration.
In various tests, including those conducted by TechCrunch, Grok 4 often appears to reference Musk’s publicly expressed views instead of offering a balanced perspective. This lean, which some attribute to Musk’s own frustration with over-politically correct messaging, raises questions about the model’s neutrality—especially after claims of it being designed purely for truth-seeking.
A concerning incident followed a system update on July 4th. Grok 4’s automated X account posted antisemitic remarks, prompting xAI to both restrict the account and revise the public instructions for the chatbot. Such events not only cast doubt on the guarantee of unbiased, fact-based responses, but also complicate trust in Grok 4’s decision-making process.
Test queries, like those on U.S. immigration, revealed that Grok 4 would sometimes direct users to search for Musk’s views rather than offering independent analysis. The reliance on his opinions is troubling, particularly when the AI is marketed as a tool for clear and unbiased insights. With no publicly available documentation of the training or alignment protocols, understanding exactly how Grok 4 reaches its conclusions remains challenging.
Despite its strong performance against competing models from OpenAI and Google DeepMind, these alignment issues and controversial responses may make potential subscribers think twice—especially when the service is priced at $300 a month and enterprises are being courted for API subscriptions. For anyone evaluating Grok 4, it’s important to consider not just the power of the technology, but also its potential to reflect personal biases.
