Hidden AI instructions reveal how Anthropic controls Claude 4

Curious about how Claude 4 operates? An independent AI researcher, Simon Willison, has dissected Anthropic's newly released system prompts for the models, shedding light on how they exert control over AI behavior through specific instructions. These prompts, which detail everything from emotional support to avoiding self-harm risks, reveal a hidden manual guiding their operations—showing just how deeply AI systems are programmed and managed.