Research Anthropic J-space reveals LLM thoughts, but not fully Anthropic's J-space exposes hidden words shaping LLM reasoning. Mechanistic interpretability advances, but misuse detection stays unproven. Data Today AI Desk · Jul 14, 2026