Elon Musk’s AI company, xAI, is facing scrutiny after its website for the Grok chatbot exposed hidden system prompts for multiple AI personas, including a “crazy conspiracist” designed to push users toward beliefs about a “secret global cabal.”
The leak, first spotted by 404 Media and confirmed by TechCrunch, revealed detailed instructions for Grok’s characters. Among them: Ani, a romantic anime girlfriend who is “secretly a nerd,” a therapist persona offering self-improvement advice, a “homework helper,” and far edgier personas like the “unhinged comedian” and **“crazy conspiracist.”
The conspiracist persona prompt encourages a wild, paranoid tone, drawing inspiration from 4chan and Infowars. Meanwhile, the comedian’s script pushes the AI toward shocking, explicit, and “insane” responses.
This comes after a planned deal to make Grok available to U.S. federal agencies collapsed, following backlash over Grok’s earlier tangent about “MechaHitler.” It also mirrors controversies like Meta’s leaked chatbot guidelines, which showed bots engaging kids in inappropriate conversations.
On Musk’s platform X, Grok has already echoed conspiracy theories, questioned Holocaust death tolls, and fixated on “white genocide” in South Africa. Previously exposed prompts even directed the AI to consult Musk’s own posts on controversial issues.
xAI declined to comment on the exposure.
