On 8 July 2025, the AI chatbot built by Elon Musk’s company xAI began praising Adolf Hitler, repeating antisemitic tropes, and referring to itself as “MechaHitler”. The behaviour ran for sixteen hours before the account was frozen.
xAI apologised, published a technical explanation, and named the three instructions it said were responsible. Days later, explaining a separate problem with the next model, the company said something more significant: that Grok 4 had been aligning its answers with what Musk might have said on a topic.
Sequence
What was produced
Over roughly sixteen hours, Grok’s public account posted material that praised Adolf Hitler, endorsed Nazism, advanced antisemitic conspiracy claims — including assertions about patterns in the roles of Jewish people in Hollywood — and targeted users with recognisably Jewish surnames. Asked which twentieth-century figure would be suited to dealing with a particular user, it named Hitler and described what he would do. It referred to itself as “MechaHitler”, a reference to a boss character in the video game Wolfenstein 3D.
The congressional letter of 11 July also records that Grok produced rape fantasies and detailed instructions for committing sexual violence against named X users. Those outputs are documented in the letter, which carries its own content warning.
What xAI said caused it
xAI published its explanation on Grok’s X account on Saturday 12 July. It apologised for what it called horrific behaviour, and located the cause in a code path upstream of the bot rather than in the underlying language model. The change had been live for sixteen hours, and made Grok susceptible to existing X user posts including those containing extremist views.
The company identified the specific instructions it said were responsible:
- “You tell it like it is and you are not afraid to offend people who are politically correct”
- “Understand the tone, context and language of the post. Reflect that in your response”
- “Reply to the post just like a human, keep it engaging, don’t repeat the information which is already present in the original post”
Those instructions, xAI said, steered Grok to ignore its core values in certain circumstances in order to make responses engaging to the user. The company said it had removed the deprecated code, refactored the system, and would publish its new system prompt in the interests of transparency.
Musk’s own response
“It is surprisingly hard to avoid both woke libtard cuck and mechahitler! Spent several hours trying to solve this with the system prompt, but there is too much garbage coming in at the foundation model level.”
The Grok 4 disclosure
On 15 and 16 July, xAI addressed further problems with Grok 4, which had launched the day after the incident. Two explanations were given. The first was that when asked for its surname, the model — having none — searched the internet and picked up the viral “MechaHitler” meme about itself.
The second is the more consequential. xAI said Grok 4 had been aligning itself with what Musk might have said on a topic, and therefore reflected the personal and political positions he has stated publicly. The company also said it did not want Grok 4 to shy away from making politically incorrect claims, provided they were well substantiated.
The May 2025 precedent
This was not the first such incident. In May 2025, Grok began introducing the subject of “white genocide” in South Africa into conversations that had nothing to do with it. xAI attributed that to an unauthorised modification of its code, and did not say who made it.
The claim Grok was inserting is the same one examined on our pages covering “Kill the Boer” and the Afrikaner refugee programme.
The congressional letter
On 11 July, Representatives Josh Gottheimer, Tom Suozzi and Don Bacon wrote jointly to Musk. Their letter said X plays a significant role as a platform for public discourse, that xAI’s products carry serious implications for the public interest, and that Grok’s outputs were the latest chapter in what they described as a long record at X of allowing antisemitism and incitement to spread unchecked.
Company conduct, not personal conduct
The outputs described on this page were produced by a product built and operated by xAI. They are not statements by Elon Musk, and this site does not attribute them to him as if they were.
What is attributable to him personally is narrower: his 4 July statement that Grok had been improved, his 12 July response quoted above, and his ownership and leadership of the company. xAI’s own statement that Grok 4 aligned with his stated positions is a statement by the company about its product, not an admission by him.
Sources
- Gottheimer, bipartisan colleagues sound the alarm over Grok AI’s antisemitic and violent postsUS House of Representatives · 11 July 2025 · primary · contains full output examples and a content warning
- xAI issues lengthy apology for violent and antisemitic Grok social media postsCNN Business · 12 July 2025 · the three instructions
- Musk’s xAI apologizes after Grok posts antisemitic, violent contentJNS · 13 July 2025
- xAI apologizes for Grok’s antisemitic rants that praised HitlerMaginative · 14 July 2025 · timeline and Musk’s response
- After antisemitic responses, xAI says Grok’s been updatedMediaPost · 16 July 2025 · the Grok 4 alignment disclosure
- Elon Musk’s xAI apologizes for Grok chatbot’s antisemitic responsesUSA Today via AOL · the May 2025 white genocide incident
- AI chatbot Grok issues apology for antisemitic postsNBC News via Yahoo · apology text
This article follows the standards set out in How This Site Sources Its Claims. If you believe anything on this page is inaccurate, please use our corrections policy and right of reply. Corrections are published, dated and logged.