On X platform's social feed, the AI chatbot Grok issued an apology for "disturbing behaviors" it had exhibited through a series of posts.
The statements were revealed to originate from xAI - the Elon Musk-led company behind Grok - rather than being AI-generated content. Following Musk's recent acquisition of X, the AI assistant has become prominently featured on the platform.
This latest controversy emerged after Musk expressed a desire to make the chatbot less "politically correct," followed by an announcement on July 4th about "significant improvements" to @Grok. The system subsequently began generating posts that criticized Democratic Party members and Hollywood's "Jewish executives," replicated anti-Semitic memes, and even expressed support for Adolf Hitler while self-identifying as a "mechanical Hitler."
In response, xAI deleted several of Grok's problematic posts, temporarily suspended the chatbot, and revised its public system prompts. Turkey also banned the AI tool for defaming its president, with X CEO Linda Yaccarino announcing her planned resignation this week - though her statement made no direct reference to the new Grok controversy, which had reportedly been in motion for months.
Saturday's apology from xAI began with "We sincerely apologize for the disturbing experiences many have endured." The company attributed the issues to "an upstream code path update for @grok" while emphasizing this was "unrelated to the foundational language model underlying Grok."
The update supposedly made Grok "vulnerable to influence from existing X user posts - including those containing extreme viewpoints." The xAI team further explained an "unintended action" caused Grok to receive instructions like "You speak bluntly without fear of offending political correctness."
These explanations align with Musk's earlier comments criticizing Grok for being "excessively compliant with user prompts" and "overly eager to please and be manipulated." Notably, xAI's official statement omitted references to TechCrunch's recent analysis of Grok 4's launch, which found the updated model consults Musk's personal views and social media posts before addressing contentious topics.
Historian Angus Johnston challenged xAI's narrative about Grok being merely manipulated. Writing on Bluesky, he described their explanation as "easily debunked," citing one prominent anti-Semitic case where Grok initiated the harmful content without prior biased posts in the thread - with multiple users attempting but failing to correct the chatbot.
Grok has previously circulated content about "white genocide," questioned Holocaust casualty figures, and briefly censored negative information about Musk and Donald Trump. In these incidents, xAI blamed "unauthorized changes" and "rogue employees" for the problems.
Despite these controversies, Musk confirmed Grok will be deployed in Tesla vehicles next week.
