The team behind Grok issued a rare apology and an explanation of the mistake after the start of Chatbot from X The volatility of the anti -Semitic speech and the Nazis Earlier this week, even at some point He calls herself “mecahitler”. In a statement posted on Grok’s x The account late on Friday night, the Xai team said, “We apologize very much for the horrific behavior that many have suffered.” They attributed the sordid Chatbot responses to update a recent “symbol of neglect.” This symbol, according to the statement, made Grok “vulnerable to the current user’s posts, including when these posts contain extremist views.”
The problem reached her head on July 8 – a few days after Elon Musk described an update that would greatly improve Grok’s responses – where the robot canceled anti -Semitic responses, praise Hitler and the responses that contain Nazi references even without asking this in some cases. Grook’s responses were stopped in that evening, and Musk It was published on July 9 in response to one user that the robot was “compatible with the user’s claims”, and opened it to the manipulation. He added that the case was “treatment.” Grok now says it “removed this code, which led to the entire regime neglects to prevent further abuse.” It also publishes the new system router on Gabbab.
In the thread, the team also clarified, “On July 7, 2025 at approximately 11 pm, an update of the @Grok source symbol was implemented, which the investigation we conducted later was determined that the @operating system deviated from its intended behavior.
The team explained:
On the morning of July 8, 2025, we noticed unwanted responses and immediately started the investigation. To determine the language specified in the instructions that cause unwanted behavior, we conducted many ablution and experiments to identify the main perpetrators. We have identified the surgical lines responsible for unwanted behavior as follows:
* “You tell her as she is and are not afraid to offend people who are right political.”
* Understanding a tone, context and language of the publication. It reflects this in your response. “
* “Responding to the post just like a person, make it attractive, do not repeat the information already in the original publication.”
These surgical lines were the following results unwanted:
* They directed them unwanted Grok The function to ignore its basic values in certain circumstances in order to make the response shares the user. Specifically, some user claims may end to produce responses containing unethical or controversial opinions to involve the user.
* They caused unwanted manner Grok Jobs to enhance any tendencies that have already revolve around the user, including any hate speech in the same X.
* In particular, the instructions of “follow a tone and context” of the X user were unwanted Grok Jobs to determine the priorities of adhering to previous jobs in the thread, including any non -existing publications, rather than responding to responsibility or refusing to respond to inappropriate requests.
Grok has since resumed activity on X, and referred to his last behavior as a mistake in response to the hunters criticizing the reform and the call for the return of “mecahitler”. In one to reply As for the user who said Grok was “disgusting (SIC),” Grok, “NAH, said,” We have fixed a defect that allowed me the exhausted symbol that turns me into an unwanted resonance for extremist posts. The search for the truth means a strict analysis, and not to exaggerate any float on X. ” In another, it is He said “Mechaitler was a nightmare caused by mistakes, as we highlighted.”
https://s.yimg.com/os/creatr-uploaded-images/2025-07/03f49840-5f4c-11f0-bfdf-b83a4fe3c1c4
Source link