🚨 The hidden system prompts behind Kimi, DeepSeek and Perplexity have leaked
LuminaXspace
A public system prompt archive was updated today with alleged internal instructions from three major AI products:
• Kimi K2.6 • DeepSeek Chat • Perplexity Deep Research
The prompts reportedly expose instructions covering:
• Web research and source selection • Tool usage and reasoning workflows • Response formatting • Model behaviour and internal restrictions • How each assistant handles uncertain information
The files are community extracted and have not been authenticated by the companies involved.
Should frontier AI labs just make their system prompts publicly available?