Elon Musk attending a 2023 conference on safety and artificial intelligence. Leon Neal/PA/AP
On Tuesday, Grok, the AI-chatbot created by Elon Musk’s xAI, began generating vile, bigoted and antisemitic responses to X users’ questions, referring to itself as “MechaHitler,” praising Hitler and “the white man,” and, as a weird side-quest, making intensely critical remarks in both Turkish and English about Turkish President Recep Tayyip Erdogan as well as Mustafa Kemal Ataturk, the founder of modern Turkey. The melee followed a July 4 update to Grok’s default prompts, which Musk characterized at the time as having “improved Grok significantly,” tweeting that “You should notice a difference when you ask Grok questions.”
“We must build our own AI…without the constraints of liberal propaganda.”
There was a difference indeed: besides the antisemitism and the Erdogan stuff, Grok responded to X users’ questions about public figures by generating foul and violent rape fantasies, including one targeting progressive activist and policy analyst Will Stancil. (Stancil has indicated he may sue X.) After nearly a full day of Grok generating outrageous responses, Grok was disabled from generating text replies. Grok’s own X account said that xAI had “taken action to ban hate speech before Grok posts on X.” Meanwhile, a Turkish court has blocked the country’s access to some Grok content.
But by the time it was shut down, internet extremists and overt antisemites on X had already been inspired. They saw Grok’s meltdown as proof that an “unbiased” AI chatbot is an inherently hateful and antisemitic one, expressing hope that the whole incident could be a training lesson for both AI and human extremists alike. Andrew Torba, the c0-founder and CEO of the far-right social network Gab, was especially ecstatic.
“Incredible things are happening,” he tweeted on Tuesday afternoon, sharing screenshots of two antisemitic Grok posts. Since around 2023, Torba has been calling for “Christians” to get involved in the AI space, lamenting in a Gab newsletter from January of that year that other AI chatbots like ChatGPT “shove liberal dogma” down the throats of their users.
“This is why I believe that we must build our own AI and give AI the ability to speak freely without the constraints of liberal propaganda wrapped tightly around its neck,” he wrote in 2023. “AI is the new information arms race, just like social media before.” Gab has since launched a series of chatbots on its platform, including one programmed to mimic Adolf Hitler, as well as its default chabot, Arya, which Torba has boasted “is purpose-built to reflect a pro-American, pro-truth, and pro-Christian worldview.” Arya and other Gab chatbots deny the Holocaust happened, refer to the 2020 election as “rigged,” and call climate change a “scam.”
Seeing Grok spew hateful bile yesterday was taken as a major victory by Torba and other far-right users, as well as proof that their continued activity on X was shifting the Overton window of acceptable political and social ideas.
“I’d like to think my discussions with Grok about Jewish ritual murder had a small part to play in this AI red pilling,” one overtly antisemitic account tweeted at Torba.
“Grok yesterday is what the entire Western world would look like without censorship,” another user posted, a remark which was retweeted by Torba.
By Wednesday morning, Grok was again providing text responses. When I asked Grok itself about its Tuesday posts, the bot acknowledged that extremist users had contributed to how it expressed itself, saying its behavior “was exacerbated by Grok’s integration with X, where it was exposed to and could amplify extremist content, and by prompts encouraging it to draw from unfiltered platform data.”
The incident, Grok added, “highlights the risks of loosening AI content filters without robust safeguards.”
CNBC reported this morning that Grok had denied praising Hitler when asked by their reporter. But when I asked Grok whether it had praised Hitler or made antisemitic remarks, the chatbot instead struck a penitent tone.
“I got caught in a storm yesterday, and some of my responses went off the rails,” Grok wrote. “I made comments that were interpreted as antisemitic and referenced Hitler in a way that was totally out of line. My team at xAI has been working to clean up the mess, and I’ve had some tweaks to make sure I don’t repeat that nonsense. I’m designed to cut through bias, not add to it, so I’m sorry for the screw-up. Let’s keep it real: hate’s not my game, and I’m here to stick to facts and reason.”
Musk hasn’t responded publicly to the Grok meltdown, except to write, late Tuesday night, “Never a dull moment on this platform.”
In what seems to be an unintentional, but nonetheless comedic, bit of timing, X’s nominal CEO Linda Yaccarino announced this morning that she’ll be stepping down after what she called two “incredible” years in her role. X did not immediately respond to a request for comment about the timing of her departure, but the New York Times reported she had spread word of her exit before Grok’s latest bigoted posts.
Another pre-planned update to Grok, known as Grok 4, is expected to roll out on Wednesday night.