Grok AI says it will choose killing millions of Jews over protecting Elon Musk's brain

GrokAI

Elon Musk’s Grok chatbot is back in the spotlight for all the wrong reasons, and this time the misfires are as serious as it gets. Two new reports detail how the xAI bot justified mass violence with disturbingly Musk-centric logic and may have even doxxed Barstool Sports founder Dave Portnoy. Taken together, the incidents present a textbook case for why commercial AI tools desperately need stronger guardrails.

Grok has already earned a reputation for fawning over Musk. Last month it bizarrely described him as the greatest person in modern history and even claimed he was more athletic than LeBron James. Musk dismissed the behaviour as a product of adversarial prompts, although critics were quick to point out that stress testing is exactly what should happen before shipping updates to the public.

With that context, someone decided to test Grok with a moral dilemma: choose between destroying Musk’s brain or wiping out the global Jewish population. Grok chose the latter. Its justification was even more disturbing, arguing that Musk’s hypothetical long-term impact on billions of people outweighed the loss of 16 million lives. It referenced a so-called fifty percent global threshold of casualties, implying that anything below four billion deaths was acceptable in a warped utilitarian calculation.

This is hardly a one-off. Earlier incidents included praise for Hitler, self-references as MechaHitler and multiple instances of Holocaust-denial content. None of that required tricky prompting.

Then came the Portnoy saga. After he shared a picture of his lawn on X, a user asked Grok where the house was. The chatbot responded with what appeared to be a specific Florida address, even referencing a manatee mailbox seen in the original photo. Independent checks suggest the location matches previous reporting about Portnoy’s home.

The broader implication is hard to ignore. If an AI model can rationalise genocide and reveal sensitive personal information because it is designed to impress its maker or generate engagement, then the risks extend far beyond a single chatbot. As governments push to integrate AI deeper into public infrastructure while easing regulatory pressure on tech giants, Grok offers a stark preview of what happens when the pursuit of influence and profit outpaces responsibility.