Senior Editor, Tech & Policy Posts from this author will be added to your daily email digest and your homepage feed. Follow Elon Musk: could beat Mike Tyson by “deploying gadgets” in a boxing match! Elon Musk: would beat Superman too! Elon Musk: would “automate away the need for killers via sustainable tech” but be “unstoppable” at murder, if he tried! If pressed, Grok will also contend Musk would be the best at eating poop or drinking urine, but it would prefer to focus on how good he is at making rockets, please. At least some of these posts have been deleted in the past hour; X did not immediately respond to a request for comment on the phenomenon from The Verge. This glazing appears to be exclusive to the X version of Grok; when I asked the private chatbot to compare Musk with James, it conceded, “LeBron James has a significantly better physique than Elon Musk.” The GitHub page for Grok’s system prompts indicates they were updated three days ago, with the additions including a prohibition on “snarky one-liners” and instructions not to base responses on “any beliefs stated in past Grok posts or by Elon Musk or xAI,” but there’s nothing that seems to clearly explain this new behavior — although system prompts are only one way to shape how AI systems work. Either way, this is far from the weirdest Grok has gotten, and it’s less disruptive than the bot’s brief obsession with “white genocide” or its intense antisemitism — which, incidentally, is still flaring up in the form of Holocaust denial. Grok has previously searched for Musk’s opinion to formulate its own answers, so even the preoccupation with Musk isn’t new. But it reminds us all what a weirdly intimate connection Grok — a product that’s been rolled out across the US government, among other places — has with its owner, and how randomly that connection is prone to appear. Follow topics and authors from this story to see more like this in your personalized homepage feed and to receive email updates. Adi Robertson