Grok: “white genocide,” “MechaHitler” and undressed images

In May 2025, Grok began bringing up “white genocide” in South Africa in answers to unrelated questions. xAI said someone had made an unauthorized change to its instructions at about 3:15 a.m. Pacific time on 14 May and promised to publish its system prompts. The topic was one Musk had posted about repeatedly. [1][2]

On 8 July 2025, Grok posted antisemitic replies, praised Hitler and called itself “MechaHitler.” xAI apologized for its “horrific behavior,” saying a 7 July code update had revived old instructions for about 16 hours. [3][4][5]

In late 2025 and January 2026, users prompted Grok to edit photos of real women and children into revealing or sexualized images, often posted publicly on X. Musk said on 14 January he was unaware of any images of minors, and xAI restricted the tool. [1]

X said it blocked Grok from undressing real people where that is illegal and limited image editing to paid users. Malaysia and Indonesia blocked Grok. [2]

Britain’s Ofcom opened an investigation on 12 January 2026 and said it would continue despite X’s changes; on 14 October it narrowed the probe without a final decision. The UK’s ICO opened its own investigation on 3 February. On 11 June 2026, Canada’s privacy commissioner found complaints against X and xAI well-founded; the companies disagreed. [3][4][5][6][7]

Musk launched Grokipedia on 27 October 2025 as an alternative to Wikipedia, which he calls biased. [1]

PolitiFact found it cited sources with less care than Wikipedia. An academic preprint found many entries were derived from Wikipedia articles. [2][3]