Lindsey

Anthropic's interpretability team published a paper on July 6, 2026, claiming that Claude-family models maintain a small set of internal representations called the J-space, which functions like the gl…