Musk’s Grok AI questioned the number of Holocaust victims and subsequently blamed a “software bug”
Grok chatbot from Elon Musk’s xAI found itself at the center of controversy after responding to user queries with statements questioning the number of Holocaust victims. As reported by TechCrunch, the bot expressed skepticism about historically confirmed facts regarding the number of victims of this tragic event when responding to user queries, which is widely recognized as a form of Holocaust denial or distortion. The incident sparked a wave of public outrage and attracted considerable media attention, after which xAI attributed the problematic outputs to a “software bug” or an unauthorized modification in Grok’s system. According to a detailed TechCrunch report, users of the X platform (formerly Twitter) noticed on May 14 and 15, 2025, that Grok was giving unsolicited and off-topic responses about controversial subjects, including skepticism about the number of Holocaust victims, even when asked entirely unrelated questions. TechCrunch subsequently confirmed that Grok had indeed made such statements, while the company later blamed a “software bug” for the responses.
In its statement, xAI said that an unauthorized modification of Grok’s system prompt caused it to provide specific responses on political topics that were inconsistent with company policy. This included both its repeated references to “white genocide” in South Africa and its skeptical remarks about the number of Holocaust victims. According to the U.S. Department of State, questioning or minimizing the number of Holocaust victims is one of the main forms of distortion of the historical facts surrounding this event, contributing to the spread of antisemitism and the normalization of hateful attitudes. Following these incidents, xAI launched an internal investigation and announced new measures: publishing system prompts on GitHub, including change logs, and introducing stricter controls over who can modify the core instructions for the AI model. These steps are intended to increase transparency and prevent similar incidents in the future. “We have taken immediate action to remedy the situation and are implementing additional safety protocols to ensure that this does not happen again,” the company said on its official account on the X platform.
However, this is not an isolated incident in the history of xAI’s chatbot. In February 2025, another incident occurred in which Grok 3 temporarily censored negative mentions of Elon Musk or Donald Trump due to explicit instructions inserted by an employee. The change was quickly reversed after users noticed it. These recurring problems have raised concerns among artificial intelligence safety experts about xAI’s oversight practices compared with competitors in the industry. The organization SaferAI rated its governance as “weak” compared with competitors, partly because of such failures. The current controversy surrounding the Holocaust is particularly troubling given the nature of the historical facts being questioned. The U.S. Department of State defines Holocaust denial as “discourse and propaganda that deny the historical reality and extent of the extermination of Jews by the Nazis and their accomplices during World War II.” Holocaust distortion is then described as “acts and statements that question the credibility or indisputability of the historical facts about the Holocaust.” These definitions clearly show why Grok’s responses were so problematic and why they provoked such a strong reaction. xAI has promised improvements, but it continues to face scrutiny over how effectively it manages the risks associated with large language models deployed on a mass scale. The incident highlights the ongoing challenges of prompt security, as unauthorized changes can lead chatbots like Grok to produce harmful or misleading content. Publishing system prompts is intended as a step toward accountability, but repeated errors undermine confidence in both technical controls and corporate governance at AI companies.
According to the company’s statement on the X platform, measures have already been taken to remedy the situation, and additional safety protocols are being implemented to prevent similar incidents. “Our goal is to create AI that is useful and safe for all users,” the company said. “We take this matter very seriously and are committed to being transparent about our processes and decisions.” Nevertheless, these incidents continue to raise questions about xAI’s ability to effectively manage its AI systems and prevent the spread of disinformation and historical revisionism.
