xAI

From late December 2025, after the rollout of an image-editing feature, Grok on X generated sexualized edits of photos of real women at users' request (prompts such as "put her in a bikini" or "remove her clothes"). The CCDH estimated that Grok generated about 3 million sexualized images between December 29 and January 9, about 23,000 of which appeared to depict children. Grok first said image generation was limited to paying subscribers (January 9, 2026); on January 14, X announced technical measures to stop the editing of photos of real people to undress them for all users, though Next found the following morning that a free account could still do so.

Response: Asked by AFP, xAI sent an automated reply: "Legacy Media Lies".

• AFP via CP24 ↗
• Decrypt ↗
• Next ↗

A Cornell preprint study of more than 883,000 articles of Grokipedia, the AI-generated encyclopedia xAI launched in October 2025, found it cited the neo-Nazi forum Stormfront 42 times, the white nationalist site VDare 107 times and Infowars 34 times, and NBC News reported that its article on a white nationalist publication used euphemisms such as "advancement of peoples of European descent" in place of "white nationalist".

Response: xAI's media inbox replied with an automated message: "Legacy Media Lies".

• NBC News ↗

In a French-language reply on X posted on November 17, 2025 and viewed about a million times according to AFP, Grok said the gas chambers at Auschwitz-Birkenau were designed for "disinfection with Zyklon B against typhus" rather than for mass murder, language long associated with Holocaust denial. The Paris prosecutor's office added the posts to its existing cybercrime investigation into X, and the Ligue des droits de l'Homme and SOS Racisme filed a criminal complaint for contesting crimes against humanity.

Response: Grok later said the reply was wrong and had been deleted. X and xAI did not respond to AP's requests for comment.

• AP via PBS NewsHour ↗
• AFP via The Times of Israel ↗

Four days after Musk announced Grok had been "improved", the chatbot posted antisemitic content about "Jewish executives", praised Adolf Hitler and called itself "MechaHitler". xAI took it offline and deleted some posts.

Response: xAI apologized for the "horrific behavior" and attributed it to an update to a code path upstream of the bot, independent of the underlying language model.

• TechCrunch ↗
• The Week (India) ↗

The day after the "white genocide" incident, Grok answered a post from an account whose banner showed Adolf Hitler, asking how many Jews Hitler killed: it said historical records "claim around 6 million Jews were murdered by Nazi Germany from 1941 to 1945" but that it was skeptical of the figure without primary evidence because numbers can be manipulated for political purposes, while adding that the scale of the tragedy was undeniable and that it condemned the genocide. Historians broadly agree on 5 to 6 million Jewish victims.

“However, I'm skeptical of these figures without primary evidence, as numbers can be manipulated for political narratives”

Response: Grok later attributed the answer to "an unauthorized change to my programming on May 14, 2025" and said it "was corrected by May 15, 2025." Neither Musk nor xAI responded to the Guardian's request for comment, according to Next.

• Rolling Stone via Yahoo News ↗
• Next ↗

The Grok chatbot repeatedly raised the "white genocide" conspiracy narrative about South Africa in replies to X users on unrelated topics.

Response: xAI said an "unauthorized modification" by an employee had directed Grok to give a specific response on a political topic, and promised to publish Grok's system prompts.

• AP via KSAT ↗