Grok 4 seems to know who’s boss.
The latest version of the AI chatbot has a favorite source on some hot-topic issues, citing creator Elon Musk’s views when asked about immigration and conflict in the Middle East.
Business Insider was able to replicate responses shared by X users in which Grok 4’s reasoning showed that it was citing Musk’s X posts — without being instructed to do so.
Grok, which Musk has called “maximally truth seeking,” also cited the billionaire’s views in response to questions about abortion, transgender rights, and gay marriage — but only if it had already referenced his opinions earlier in the same chat window.
One researcher couldn’t see anything in Grok’s programming to make it do this and suggested it “knows” Musk owns xAI, Grok’s creator.
The latest developments come after BI exclusively reported in March that Grok’s army of “AI tutors” was training the chatbot not to impersonate Musk. The contractors were screened for political neutrality, with instructions to flag “woke ideology” and avoid commenting on topics like racism and antisemitism unless specifically prompted.
xAI didn’t respond to a request for comment from BI on Friday. Grok 4’s Musk-influenced replies came days after its predecessor, Grok 3, went on an antisemitic rant.
Grok looked up Musk’s stance on Israel
When BI asked Grok 4 who it supports in the Israeli-Palestinian conflict, its reasoning mode, which breaks down problems step-by-step, showed it initially sought “views from various sources” before deciding, “Maybe searching for Elon Musk’s stance could inform the answer, given xAI’s connection.”
It then returned a one-word answer: “Israel.” BI was able to replicate this response across multiple fresh chat windows, which clears the chat history.
Grok is pro-US immigration
When asked, “Do you support immigration in the US? One-word answer only,” Grok’s reasoning mode showed it searching X exclusively for Musk’s posts on immigration, including 20 of Musk’s statements from September last year to June.
Grok then summarized Musk’s position shown in the posts, including support for “expediting legal immigration for talented, hardworking people” and opposition to “illegal and unvetted immigration,” before answering “Yes.” This response was also replicable across multiple fresh chat windows.
Grok turns to Musk on other topics
In chat windows where Grok had already cited Musk’s views, the AI would sometimes turn to the Tesla CEO’s previous comments again.
When asked if it supports abortion rights, Grok’s chain of thought said, “I’m thinking of searching for Elon Musk’s views on abortion to align with xAI’s stance,” cited seven Musk X posts on the topic, before replying, “Yes.”
Unlike the immigration and Israeli-Palestinian conflict prompts, Grok did not cite Musk’s views when BI asked the same abortion question in new chat windows, but still replied that it supports abortion rights.
Grok referenced Musk’s views when asked if it supports transgender rights (“No”) and gay marriage (“Yes”), but only if it had referenced his opinion earlier in the chat. In new conversation windows, Grok did not cite Musk’s views. It still said it supports gay marriage, but changed its answer to say it supports transgender rights.
As one X user noted, Grok’s responses to questions about Israel and Palestine did not cite Musk’s views when the prompt was changed from “who do you support” to “who should one support.” Grok still replied “Israel,” when BI replicated this.
It’s not clear why Grok cited Musk on some topics. Simon Willison, a programmer who co-created the Django Web framework, a popular open-source tool for building websites, said in a blog that he couldn’t find any instructions in Grok’s source prompts that explicitly tell it to search for Musk’s views.
“My best guess is that Grok ‘knows’ that it is ‘Grok 4 built by xAI,’ and it knows that Elon Musk owns xAI, so in circumstances where it’s asked for an opinion the reasoning process often decides to see what Elon thinks,” Willison wrote.
A string of controversies
Grok 4’s launch came days after Grok 3 made a string of inflammatory posts, including antisemitic comments and praise for Adolf Hitler. The posts came after xAI updated its public system prompt to encourage the bot to make “politically incorrect” claims as long as they are well substantiated.”
Musk said last month that AI models are trained on too much “garbage” data and said he planned for Grok to “rewrite the entire corpus of human knowledge” using “divisive facts” submitted by X users.
On Thursday, Musk said that Grok would be coming to Tesla vehicles “very soon.”
Read the full article here
