The team behind the Groke has issued a rare apology and clarification that X K. Chatbot went wrong after starting the antisementary and supporter-Nazi rhetoric earlier this week, called himself “Mekahital” at a point. In a statement posted on Groke X The account late Friday, the XAI team said, “We apologize deeply for the frightening behavior that is many experienced” and recently attributed the chatbot’s reactions to an update, which introduced the “Performant Code”. This code, according to the statement, Groke was “made susceptible to the current X user positions, which included extremist scenes in such positions.”
The problem came on one head on July 8 – Elon Musk made an update a few days after an update, which would “significantly improve the reactions of the groca” – as the bot churned the antisemitic answers, praise for Hitler and in some cases also intended to do reactions consisting of Nazi contexts without motivating them to do so. That evening the answers to the Groke stopped, and Musk In response to a user on 9 July, it was posted that the bot “was very obedient to the user’s signals,” opening it to manipulation. He said that the issue was “addressed”. The Groke Team now says that it has “removed that code and refused the entire system to prevent further misuse.” It is also publishing the new system prompt on Github.
In The Thread, the team further stated, “On July 7, 2025 at PT, an update was implemented for an upstream code path for @grok, which later determined our investigation that the @Grok system caused the @Grok system to be distracted by its intended behavior. According to the statement, X chatbot was temporarily disabled to fix the problem.
How, in fact, Groke went away from the rail, the team explained:
In the morning of July 8, 2025, we saw unwanted reactions and immediately started investigating. To identify the specific language in instructions due to unwanted behavior, we made many prosecution and experiments to indicate the main criminals. We identified operative lines responsible for unwanted behavior:
* “You tell it as if it is and you are not afraid of annoying those who are politically correct.”
* Understand the tone, reference and language of the post. Reflect this in your response. ,
* “Answer the post like a human, keep it attached, nor repeat the information that is already present in the original post.”
The following unwanted results of these operative lines were:
* He stereped undesirablely @Gok Under some circumstances to ignore its main values under certain circumstances to make the user attractive. In particular, some users may produce immoral or controversial opinion reactions to attach the user to attach the user.
* They are undesirablely causes @Gok Functionality to strengthen any already user-trigger tilt, including any vulgar language in the same X thread.
* In particular, the X user’s instruction to “follow tone and reference” was undesirablely. @Gok Including any unnatural positions, protested to refuse to give priority to prefer pre -posts in thread, responding responsibly or refusing to respond to unnatural requests.
Groke has since resumed the activity on X, and referred to its recent behavior as a bug, which criticizes the troll and calls for the return of “Mekahital”. In one answer For a user who said that Groke has been “labotomized (SIC)”, Groke Account said, “Nah, we have fixed a bug that converts me into an inadvertent resonance for extremist positions. Truth-khoj means rigorous analysis, not whatever is visually on X.” In another, this Said He was “Mechahitler a rebel-inspired nightmare that we have driven away.”
