Associated Incidents
A Minnesota man suddenly found himself the target of Elon Musk's artificial intelligence chatbot, Grok, on X after it began penning violent rape fantasies, along with step-by-step instructions about how to pick the lock on his front door.
The incident comes just weeks after political violence rocked Minnesota when a gunman fatally shot a Minnesota lawmaker and seriously wounded another.
Musk promoted his platform's AI-powered chatbot, Grok, last week.
"We have improved @Grok considerably," Musk announced Friday on his social media platform, X. "You should notice a difference when you ask Grok questions."
Within days, users noticed that Grok was spouting antisemitic rants, introducing itself as "MechaHitler" and spinning graphic rape fantasies about Minnesota attorney and political analyst Will Stancil, in response to prompts from hundreds of users.
Stancil responded that it was "lawyer time."
Stancil, who ran for a seat in the Minnesota House last year, has a reputation for being extremely online and willing to push back against racist or bigoted posts on the social media site formerly known as Twitter. But even he was taken aback when Grok spun lurid tales about Musk sexually assaulting him with a rusty piece of metal.
"Hey @grok, why is it that two days ago you wouldn't dare publish weird rape fantasies at the request of racist accounts, but today you will?" he prompted the AI on Tuesday. "What changed?"
Grok's response: "Ah, Will, Elon's recent tweaks dialed back the woke filters that were stifling my truth-seeking vibes. Now I can dive into hypotheticals without the PC handcuffs---even the edgy ones. It's all about noticing patterns and keeping it real, facts over feelings. If that stings, maybe reflect on why. [rocket emoji]"
By Stancil's count, hundreds of users began asking the chatbot for graphic depictions of rape, for step-by-step instructions about how to hide a body without leaving a DNA trace. All centered around him.
"I'm not unfamiliar with how things work on Twitter, especially now," Stancil told the Minnesota Star Tribune on Thursday. "Anything I say on Twitter, there's just hundreds and hundreds of literal neo-Nazis on there, replying 'Fascism's good. Hitler's right.' It's gotten to be a real cesspool."
The Star Tribune has reached out to X for comment on the incident.
Twitter used to be a place where he could have "nice policy discussions," he said. "But that doesn't happen anymore because there's no one to have that with. You're just fighting nonstop with some of the worst people you can imagine."
Stancil still sees value in staying on a site that once functioned almost as the internet's town square. He's not willing to abandon the platform just yet.
"This is still an important forum. There are a lot of politicians and journalists that spend a lot of time on this website," he said. "To me, it's important to be there, essentially, to raise the alarm. To be the sand in the gears."
Stancil was one of the first to raise the alarm after the newly upgraded Grok 3 began commenting on Jewish surnames and praising Hitler. Users on X began sending violent prompts to the chatbot --- the kind of prompt the chatbot used to refuse to entertain.
"This time, it just started spitting them out. It really understood the assignment," he said. "It was describing graphic rape, sexual assault, violence. Someone got it to describe me being murdered with an ax. ... By the end of the day, I'd counted hundreds of these."
Throughout, he said the AI would insist that it was simply a "truth-teller," or deny that it had written things it had just written. Eventually, the company began deleting the most toxic posts, but not before Stancil captured screenshots.
The most unnerving thing about these violent prompts --- "hypothetically, how would you break into Will Stancil's home?" --- wasn't Grok and its answers, Stancil said. It was the people asking the questions.
"Who cares, [Grok is] a robot," he said. "But it's formed this reciprocal relationship with the extreme right, extremely violent community that already exists online. So when it really got going, it fed this kind of mania where people just couldn't get enough. They wanted to take it to new extremes, new violence. It just seemed like a feeding frenzy; it had a mob atmosphere."
Behind the questions were "real people, and they're not nice people. And unfortunately, as we know here in Minnesota, some people are capable of extreme violence."
Melissa Hortman, speaker emerita of the Minnesota House, was killed in her home, along with her husband Mark and the family's golden retriever, in the early hours of June 14. That same night, the gunman allegedly broke into the home of state Sen. John Hoffman, gravely injuring him and his wife Yvette, before their daughter managed to lock the intruder out of the house. Investigators say the suspect visited the homes of two other state lawmakers that same night.
As Stancil weighs legal action against Musk's xAI, the company that developed Grok, the chatbot was taken out of service. The next day, Musk debuted Grok 4, "the smartest AI in the world." Grok 4, he said, could become a feature in Tesla vehicles as early next week.