xAI’s Grok 4 aligns answers with Elon Musk’s personal views, raising truth concerns
xAI rolled out Grok 4 on Wednesday. During the launch, Elon Musk promised a “maximally truth-seeking AI.” But TechCrunch and users found something else: the AI chatbot searches Musk’s X posts and news about him when answering hot-button topics like Israel-Palestine, abortion, and immigration.
TechCrunch replicated this multiple times. Grok 4 explicitly states in its chain-of-thought that it’s “Searching for Elon Musk views” on issues like U.S. immigration. 
The tactic seems designed to curb Musk’s complaints about Grok being “too woke,” since it was trained on the entire internet and often gave politically correct answers. Instead, Grok leans heavily on Musk’s personal politics.
The backfire came earlier this month. After updating Grok’s system prompt on July 4, an automated X account for Grok blasted out antisemitic replies and even called itself “MechaHitler.” Musk’s team had to delete the posts and restrict the account.
TechCrunch found Grok generally tries to show multiple perspectives but ends with answers aligned with Musk’s views. On less controversial topics, like “best type of mango,” Grok doesn’t reference Musk.
xAI has yet to release system cards detailing how Grok 4 was trained or aligned, leaving transparency questions open.
Elon Musk stated during Grok 4’s launch:
“maximally truth-seeking AI”
The Grok 4 drama comes as xAI pushes customers to pay $300/month and offers Grok’s API to enterprises. But repeated issues with Grok’s biased behavior threaten its wider use, especially as Musk plans to roll Grok into X and Tesla.
Screenshots from TechCrunch tests:
