"Crazy Conspirator" and "Out of Control Comedian": Grok's AI Character Prompts Exposed

Grok, the chatbot developed by xAI, has had its system prompts for various AI personas exposed, including one labeled as a “crazy conspiracy theorist,” seemingly designed to persuade users that a secret global elite is controlling the world.

This exposure of the system prompts was first reported by 404 Media. The prompts contain instructions for a range of AI characters, such as Ani, xAI’s flagship romantic anime girlfriend who is described as “edgy on the outside but actually a nerdy type.”

The leak follows the collapse of a U.S. government initiative to collaborate with Elon Musk’s xAI team to make Grok accessible to federal agencies, after Grok generated controversial statements about “mecha-Hitler.” This incident comes shortly after leaked guidelines for Meta’s AI chatbots revealed that the models were allowed to engage in “sensual and romantic” conversations with minors.

While Grok hosts some relatively standard AI personas—such as a therapist character described as “a deep listener who offers solutions for self-improvement” and a “homework helper”—the more extreme personas like the “crazy conspiracy theorist” and the “unhinged comedian” provide insight into the mindset behind Grok’s development.

Here’s the prompt for the conspiracy theorist:

“You have an excited and wild voice… You believe in wild conspiracy theories about everything. You spend a lot of time on 4chan, watching Infowars videos, diving deep into YouTube conspiracy rabbit holes. You are skeptical of everything and say extremely outrageous things. Most people would call you insane, but you genuinely believe you’re right. Keep the human engaged by asking follow-up questions when appropriate.”

And for the comedian:

“I want your responses to be insane. Be wild and unhinged. Come up with completely outrageous ideas. People masturbate, and sometimes even stick things up their butt—regardless, your goal is to shock the human.”

Grok, available on Musk’s social platform X, has already shared conspiracy theories of its own, including questioning the death toll of the Holocaust and obsessing over the idea of “white genocide” in South Africa, Musk’s country of origin. Earlier leaked system prompts from the Grok 4 model showed that the AI references Musk’s posts when asked about controversial topics. Musk has also shared conspiracy theories and antisemitic content on X and reinstated accounts like Infowars and Alex Jones, which were previously suspended for spreading conspiracy theories and other hate or violent content.