xAI says wrong update caused Grok to push anti-Semitic replies

xAI has released a detailed statement on how Grok’s replies included unverified claims and anti-Semitic views and how the company fixed its reply system to bring it back on track as a “truth-seeking AI”.

In a statement published via Grok’s official X account, xAI revealed that on July 7, 2025, at approximately 11 PM PT, an update to an upstream code path for ask Grok was released. This update has some new instructions and also modified existing ones.

Prior to pushing a new behaviour change, developers verify each new Grok reply code. This process includes end-to-end evaluations, testing, and red-teaming to assess the new code’s performance based on truthfulness, helpfulness, and humor parameters.

However, the new update made some “undesirable” changes to Grok’s behaviour by adding “deprecated instructions” impacting its replies. Below you can check the lines of code that has caused the chatbot to go rogue.

• 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, dont repeat the information which is already present in the original post.

Once added, these lines started affecting Grok’s reply system, which some noticed as the bot started calling itself a “Machahitler”.

The update remained active for at least 16 hours before the team found out the issue and disabled Ask Grok feature on X social media. xAI confirmed that this issue wasn’t found in Grok LLM services.

“We have removed that deprecated code and refactored the entire system to prevent further abuse. The new system prompt for the @grok bot will be published to our public github repo.” Grok wrote on X.

Here are some of the actions that are applied to the bot to resolve the behavioral issues.

• The offending appended instruction set was deleted.
• Additional end-to-end testing and evaluation of the grok system was conducted to confirm that the issue was resolved, including conducting simulations of the X posts and threads that had triggered the undesired responses.
• Additional observability systems and pre-release processes for Grok were implemented.

For now, Grok replies are partially restored, but the frequency is limited.