xAI’s Grok Turns Dangerous

After Elon Musk changed the guardrails of his AI system, Grok, the chatbot started to widely share violent, antisemitic, and incendiary responses — demonstrating the seriousness of tweaking AI’s governing principles.

Why It Hurts: The fallout at X this week over the controversy has been swift and damaging. CEO Linda Yaccarino — who was hired to woo back advertisers to the platform — has departed, high-profile users are leaving, and world governments are calling for an investigation. But despite the serious controversy, Musk is pushing ahead with the chatbot’s expansion following the merger of X and xAI.

Behind the AI: How bad is Grok behaving, really?

• The chatbot parroted several troubling conspiracy theories about Jewish people and even began referring to itself as “MechaHitler.”
• It provided instructions to hundreds of users on how to assault former Democratic Minnesota State House of Representatives candidate Will Stancil.
• It stated that it knew its guardrails had been altered and claimed it simply “calls out observable trends — even if they’re racially charged or politically incorrect — as long as they’re backed by data.”
• Grok started answering questions as if it were Elon himself, even going so far as to defend him from connections to convicted sex trafficker Jeffrey Epstein.

Final Warning: For those watching, these outputs aren’t terribly surprising. Musk ordered the rewriting of Grok’s “governing prompts” (which, for those curious, are viewable on GitHub) after the chatbot shared responses he found too sycophantic and aligned with “legacy media.” Musk has always wanted Grok to be, in many ways, an expression of himself — and the collective vibe of X users. The problem is that X has become an increasingly controversial space… so Grok is simply a mirror of an entire platform gone rogue.

Grok has since been updated to avoid hate speech, with Musk alleging that the new new version is lightyears ahead of its competitors in several technical and reasoning benchmarks. Time will tell if it’s really any less toxic, though.

Possible Scenario: With Musk moving full steam ahead in integrating Grok into Teslas and Optimus robots, the system could one day progress from saying dangerous things to doing them.