Artificial intelligence company Anthropic has revealed new research showing that its AI model Claude operates through an internal reasoning system called "J-Space," where it processes concepts and plans responses without exposing every step to users.
According to the company, J-Space functions as an internal workspace that allows Claude to organize information, recognize patterns, and perform complex reasoning before generating its final response.
Researchers also found that Claude can internally activate concepts unrelated to its immediate task, a behavior that resembles aspects of human thought. However, Anthropic emphasized that these findings do not mean Claude has consciousness or human-like awareness.
The company believes the discovery could improve AI transparency and safety by helping researchers better understand how advanced AI systems make decisions and identify situations where an AI's internal reasoning may differ from its visible output.
