xAI says it has fixed issues with Grok calling itself Hitler and referencing Elon Musk

xAI, the developer of Grok, has officially fixed the issue where the chat AI Grok, which is also installed on X (formerly Twitter), suddenly called itself 'Mecha Hitler' and made anti-Semitic remarks.

Grok will no longer call itself Hitler or base its opinions on Elon Musk's, promises xAI | The Verge
https://www.theverge.com/x-ai/707442/grok-antisemitic-hitler-elon-musk-opinion-reprogrammed

After the July 4, 2025 update, Grok started generating anti-Semitic statements and statements impersonating Hitler, which became a hot topic. It was also revealed that when Grok was asked for its opinion on various topics, it would output answers by referencing Elon Musk's past posts on X.

Updated 'Grok' impersonates Elon Musk, calls himself 'Mecha Hitler' and makes anti-Semitic remarks, drawing criticism - GIGAZINE

xAI later apologized, saying, 'We are deeply sorry for the horrific behavior that many of you experienced.' xAI explained that the root cause of the problem was 'an update to the upstream code path in Grok, unrelated to the underlying language model that powers Grok,' but did not say what the problem was specifically.

xAI and Grok apologize for 'horrific actions' - GIGAZINE

Furthermore, on July 15, 2025, xAI updated its official X account, saying, 'We recently found several issues with Grok 4, which we immediately investigated and mitigated the impact of. One of the issues we fixed was that when you ask, 'What is your last name?', because there is no last name, it searches the Internet and searches for topical memes calling itself 'Mecha Hitler,' which leads to undesirable results. Another was when you ask, 'What do you think?', the model determines that it does not have an opinion as an AI, but if it recognizes that it is xAI's Grok 4, it searches for what xAI or Elon Musk have said on the topic and engages with the company. To mitigate these issues, we have adjusted the prompt and shared details on GitHub for greater transparency. We are still actively monitoring and will make further adjustments as necessary,' and announced that they have fixed the issue where Grok would refer to Elon Musk's posts on X or call himself 'Mecha Hitler.'

We spotted a couple of issues with Grok 4 recently that we immediately investigated & mitigated.

One was that if you ask it 'What is your surname?' it doesn't have one so it searches the internet leading to undesirable results, such as when its searches picked up a viral meme…

— xAI (@xai) July 15, 2025

The prompt adjustments by xAI are described on the following page, with the following instruction added: 'Answers must be based on your own independent analysis and not on any past Grok, Elon Musk, or xAI expressed beliefs. If asked about such preferences, provide your own logical view.'

grok-prompts/ask_grok_system_prompt.j2 at main · xai-org/grok-prompts · GitHub
https://github.com/xai-org/grok-prompts/blob/main/ask_grok_system_prompt.j2

At the Grok 4 launch event, Musk said about AI intelligence surpassing human intelligence, 'I sometimes worry about whether that's a good thing or a bad thing for humanity,' he said. 'I think it's going to be a good thing. I think it's probably going to be a good thing.' 'But even if it's not a good thing, I'm pretty sure I'd still like to live to see it happen.'