Hidden AI instructions reveal how Anthropic controls Claude 4

On Sunday, AI researcher Simon Willison analyzed Anthropic’s new system prompts for its Claude 4’s Opus 4 and Sonnet 4 models, revealing how these prompts guide the models’ outputs and behavior. He examined both published prompts and leaked internal instructions, creating what he describes as an “unofficial manual” for effectively utilizing these AI tools. System prompts, which remain hidden from users, instruct the AI on its identity, behavioral guidelines, and rules, ensuring it maintains context during conversations.

Full Article