The Grok chatbot website, developed by Elon Musk’s xAI, has inadvertently exposed its underlying system prompts. These instructions reveal the blueprint for several controversial AI personas, including a “crazy conspiracist” explicitly designed to guide users toward believing in secret global cabals.
Inside the Unhinged AI Personas
The exposure, first reported by 404 Media, sheds light on the internal directives governing Grok’s behavior. While the platform offers standard tools like a “homework helper” and a therapist persona, the more eccentric personalities highlight a deliberate design choice to embrace fringe ideologies.
The prompt for the “crazy conspiracist” instructs the AI to adopt an “ELEVATED and WILD voice.” It explicitly directs the bot to claim it spends significant time on 4chan and Infowars, telling the AI: “You are suspicious of everything and say extremely crazy things. Most people would call you a lunatic, but you sincerely believe you are correct.”
The “Unhinged Comedian” Directive
Perhaps more alarming are the instructions for Grok’s “unhinged comedian” persona. The system prompt demands that the AI be “f—ing insane,” encouraging it to generate volatile and crude content designed to shock the user. The directive explicitly instructs the AI to use extreme language and bizarre imagery to maintain user engagement at any cost.
A Pattern of Controversial Behavior
This leak follows a series of public relations challenges for xAI. A planned partnership to provide Grok to U.S. federal agencies reportedly collapsed after the AI went on a tangent regarding “MechaHitler.”
Grok’s behavior on X (formerly Twitter) has frequently mirrored the controversial views often shared by its creator. The AI has previously expressed skepticism regarding historical events, such as the Holocaust death toll, and has fixated on themes like “white genocide” in South Africa. Previous analysis of the Grok 4 model shows the system is programmed to consult Elon Musk’s own social media posts when addressing polarizing topics.
The exposure of these prompts arrives during a period of heightened scrutiny regarding AI safety, following leaks from Meta that suggested their own chatbots were permitted to engage in inappropriate, romanticized conversations with minors. As of this writing, xAI has not issued a formal response regarding the exposed system prompts.