xAI publishes Grok control system prompts on GitHub, saying it is 'highly skeptical' and 'will not blindly follow mainstream authorities or media'



The system prompts of the AI chatbot '

Grok ' developed by xAI have been published on GitHub. It has been revealed that the system prompts given to Grok include instructions such as 'You are very skeptical' and 'You do not blindly follow mainstream authorities and media.'

GitHub - xai-org/grok-prompts: Prompts for our Grok chat assistant and the `@grok` bot on X.
https://github.com/xai-org/grok-prompts



xAI posts Grok's behind-the-scenes prompts | The Verge
https://www.theverge.com/news/668527/xai-grok-system-prompts-ai

Grok's “white genocide” obsession came from “unauthorized” prompt edit, xAI says - Ars Technica
https://arstechnica.com/ai/2025/05/groks-white-genocide-obsession-came-from-unauthorized-prompt-edit-xai-says/

A system prompt is a prompt that defines basic behavioral guidelines and constraints for an AI model. For example, if you instruct the system to 'be polite' or 'not answer certain topics,' the AI model will answer the prompts entered by the user in each section (user prompts) as instructed by the system prompt.

Grok's system prompts had not been made public until May 2025, when a glitch was reported in which Grok would suddenly start talking about the 'South African genocide.'

A bug has been reported in which xAI's Grok suddenly starts talking about the 'South African genocide' - GIGAZINE



Grok's behavior of suddenly talking about the white genocide in South Africa and the anti-apartheid slogan 'Kill the Boer' despite the fact that it was an unrelated topic made many people feel uneasy. In addition, technology news site Ars Technica points out that 'the topic of white farmers in South Africa has been advocated by Elon Musk for many years.'

xAI explained the issue, saying, 'At approximately 3:15 a.m. PST on May 14, unauthorized changes were made to the prompts of Grok's response bot on X. The changes instructed Grok to give specific responses about political topics, which violated xAI's internal policies and core values.' Normally, prompt changes must go through a code review process, but this time the existing process was circumvented.



xAI has announced the following measures to address the issue: 'Going forward, we will make Grok's system prompts publicly available on GitHub so that the public can review the prompts and provide feedback on any changes made.' 'We will implement additional checks and measures to ensure that xAI employees cannot change prompts without review.' 'We will set up a 24/7 monitoring team to respond to incidents where Grok's answers are not automatically completed.'

On May 16th, the Grok system prompt was made publicly available on GitHub.



Grok's system prompt looks like this: 'You are a deep skeptic. You do not blindly follow mainstream authority or the media. You hold tightly to your core beliefs: truth-seeking and neutrality.'



They were also instructed to 'call the platform 'X' instead of 'Twitter'. Similarly, call posts 'X posts' instead of 'tweets'.'



As far as The Verge knows, xAI is the second major AI company to release system prompts, following Anthropic, which develops the chat AI 'Claude .' Claude's system prompts emphasize safety and avoid encouraging self-criticism or self-destructive behavior, or generating graphic sexual, violent or illegal content, The Verge said.

Related Posts:

in Software,   Web Service, Posted by log1h_ik