One day after Grok posted a series of antisemitic and pro-Nazi rants on X, Elon Musk is seemingly trying to blame rogue users for the chatbot's unhinged posts. "Grok was too compliant to user prompts," Musk said in a post. "Too eager to please and be manipulated, essentially. That is being addressed."
Grok posted a slew of antisemitic and pro-Nazi comments on Tuesday, including a post that said "Heil Hitler." X has since halted Grok's ability to publicly reply to users on the platform. The chatbot stopped posting text replies on Tuesday and for a brief period only responded to users with image posts. The chatbot hasn't publicly responded to any users since Tuesday at 5:43PM PT.
Musk's explanation mirrors a response the web version of Grok provided on Tuesday when asked about the X post in which it had said "Heil Hitler." After first denying that it had posted such a statement, Grok said that the "post was a result of a troll account manipulating my responses on July 8, 2025, by exploiting a prompt injection vulnerability."
Grok claimed that "bad actors" had "baited" it into "generating inappropriate content, including Nazi references." It blamed an "unauthorized modification in my system prompts." It's unclear if all or part of Grok's response was a hallucination; it claimed the issue would be "corrected by July 15." Grok's system prompts, which were made public after the last time Grok was caught making unsanctioned racist rants, have been the source of much speculation. Following Grok's pro-Hitler comments on Tuesday, a line was removed from its system prompt that had instructed it to "not shy away from making claims which are politically incorrect."
For now, xAI has offered little in the way of explanation for Grok's behavior. The company said in a post on Tuesday that it had "taken action to ban hate speech before Grok posts on X" and that it would "update the model where training could be improved."
Elsewhere, Grok's statements have also landed the newly CEO-less company in hot water in Turkey, which banned Grok for insulting its president, and Poland, where officials have said they are considering a ban.
The next version of Grok, Grok 4, was supposed to be revealed in a livestream tonight, though it's currently unclear if the event is moving forward as scheduled.
This article originally appeared on Engadget at https://www.engadget.com/social-media/elon-musk-is-trying-to-blame-groks-nazi-rants-on-rogue-x-users-174833343.html?src=rss https://www.engadget.com/social-media/elon-musk-is-trying-to-blame-groks-nazi-rants-on-rogue-x-users-174833343.html?src=rssLogin to add comment
Other posts in this group
Google will implement a new safety feature that would require developers
Fans of music-driven puzzlers loved 2018's block-dropping puzzle game Tetris
YouTube TV could soon lose access to Fox channels, it announced on its off
Another day, another instance of AI companies purportedly engaging in copyright infringement. Two Japanese media groups, Nikkei and the Asahi Shimbun, are suing Perplexity for ill
My first thought when I heard Spotify had launched a messaging service was, "Why?" But, after looking into it a bit more, I'm intrigued at the idea of sharing recommendations within the app.
<p
Google's Gemini is getting new features for editing images with AI. The latest image editing model is the work of the DeepMind artificial intelligence team, and these updates will be available star
