Microsoft’s early beta of GitHub Copilot Chat is already generating controversy after researcher Marvin von Hagen posted the system’s confidential rules on Twitter. The prompt, which instructs the AI to “respectfully decline” if asked for its rules, was shared publicly alongside a link to the full set of guidelines.

The leak quickly climbed to the top of Hacker News, where it sparked a deep debate about the nature of LLM prompting. Commenters dissected the use of second-person address in system prompts—such as “You are an AI programming assistant”—and questioned why developers frame instructions as if speaking to an entity, rather than simply describing behavior. Some argued this anthropomorphism is a deliberate UI choice, making the system intuitive at the cost of clarity.

Others in the thread pointed out that raw models like LLaMA respond better to descriptive prefixes, while instruction-tuned models accept commands in a conversational format. The leak serves as a rare window into how a major vendor actually structures its AI guardrails, and the conversation reflects a community still wrestling with what it means to talk to a machine as if it were a person.