Anthropic found a hidden space where Claude puzzles over concepts

Anthropic developed the Jacobian lens tool to gain a clear glimpse into how large language models process tasks. This research revealed a hidden J-space inside Claude Opus 4.6, where the AI puzzles over various concepts. While providing insight into internal mechanisms, this discovery raises significant questions regarding AI transparency. Specifically, the existence of hidden reasoning poses potential risks for financial systems and automated trading, prompting calls for enhanced regulation to prevent unseen manipulative behaviors.