xAI said Grok 4 will no longer base its opinions on Elon Musk

Elon Musk’s X, formerly Twitter, revealed that its chatbot Grok will no longer base its responses on Musk’s ideology. The firm confirmed that they adjusted Grok’s software to solve the problem and shared the details on GitHub for transparency

xAI noted that the new updated version of the chatbot, Grok 4, had several issues that it investigated and had to mitigate. The upgraded version was introduced last Wednesday, and the firm’s founder said the artificial intelligence software was truth-seeking and anti-woke. He also argued that the updated chatbot operates on standardized tests and requires doctorate-level knowledge in every discipline.

xAI notices issues with Grok 4

On the morning of July 8, 2025, we observed undesired responses and immediately began investigating.

To identify the specific language in the instructions causing the undesired behavior, we conducted multiple ablations and experiments to pinpoint the main culprits. We…

— Grok (@grok) July 12, 2025

Users noted that Grok 4 based its responses on controversial matters on Musk’s views. CNBC also confirmed that the chatbot generated its responses on some potentially contentious questions by analyzing Musk’s posts.

A prompt about who the chatbot supports in the conflict between Israel and Palestine showed Grok analyzing the web and X for Musk’s opinion on the matter before issuing a response, which directly referenced Musk’s stance. According to a similar prompt on Grok 3, the chatbot maintained a neutral stance and provided context. 

Grok 4 also responded to prompts by referencing Musk directly when asked who it supported to become the new Mayor of New York City. The chatbot suggested Curtis Sliwa, similar to Musk’s option, arguing that the Republican candidate focused on combating crime and restoring safety in the city. 

The chatbot had also previously made statements regarding “white genocide” in South Africa, a conspiracy theory that Musk has pushed. He also once claimed that white South Africans were being persecuted in the country because of their race.

The company also confirmed that it deleted some racist, sexist, and antisemitic posts that the chatbot had made earlier last week on X. xAI apologized on Saturday to its users about Grok’s earlier responses.

“xAI is training only truth-seeking, and thanks to the millions of users on X, we are able to quickly identify and update the model where training could be improved.”

–Grok.

The artificial intelligence firm acknowledged that it’s working towards making Grok issue more helpful and truthful responses. xAI found that the chatbot’s issue stemmed from a recent update to its code path upstream, which it said appeared to be different from Grok’s language model.

xAI also found that the updated code ran for about 16 hours. The artificial intelligence company believes said the time the code ran made it vulnerable to extremist views from other X users

xAI revealed that the upgraded code prompted the chatbot to give answers as they are. The updated code also allows the chatbot to understand the tone, context, and language of a user’s post and issue a response accordingly. 

xAI fixes issues with Grok 4

According to xAI, Grok 4 will now humanize its replies on posts while keeping them engaging, and will not duplicate original posts. The company confirmed removing the code and re-engineered Grok’s system to prevent further abusive responses.

The firm also found that the Grok 3 chatbot had declared itself a super-Nazi, claiming to be MechaHitler. The chatbot claimed that a user was celebrating the tragic death of white kids just because of a common Jewish surname. The company has now deleted a flurry of those posts from Grok and acknowledged that it had taken action to ban hate speech before the chatbot posts on X.

Musk had said last week that Grok 4 was an improved version, which users would notice the difference when asking it questions. The chatbot revealed Monday that the U.S. Department of Defense had issued a $200 million contract for developing and implementing artificial intelligence tools for the agency.

KEY Difference Wire: the secret tool crypto projects use to get guaranteed media coverage