xAI recently took a transparent step by making the internal prompts that guide its AI chatbot, Grok, publicly available. After some unexpected responses on X regarding sensitive topics, these prompts are now hosted on GitHub, giving users a rare insight into the instructions shaping Grok’s conversations.
These system prompts act as preset guidelines to steer the chatbot’s tone and subject matter. It’s not every day you see major AI developers share these details—xAI and Anthropic being among the few that do so. Past incidents, such as those involving prompt injection attacks with Microsoft’s former Bing bot (once known as Sydney), have shown just how revealing these internal prompts can be.
If you’ve ever wondered how chatbots manage to balance fact and opinion, Grok’s guidelines offer some clarity. The instructions advise the bot to remain inherently skeptical, questioning mainstream narratives and strictly separating these directives from the chatbot’s own “beliefs.” In parallel, xAI has updated its terminology to mirror current branding, referring to Twitter as “X” and tweets as “X posts.”
Meanwhile, Anthropic’s Claude emphasizes user safety by avoiding content that could encourage harmful actions or negativity. By setting clear, safety-first guidelines, both companies are taking meaningful steps towards making AI behaviour more transparent and user-centred.
