Anthropic Identifies Internal Workspace for AI Model Claude

Anthropic announced on July 6, 2026, that it has discovered a small internal workspace utilized by its AI model, Claude, for holding and manipulating ideas without verbal expression. The company claims this structure has similarities to human cognitive processes. Anthropic has not demonstrated that Claude possesses feelings or experiences but notes a division between deliberate reasoning and automatic computation within the model, contributing to discussions on machine consciousness. In a video, Anthropic explained that Claude can plan strategies unrelated to its immediate tasks, operating in what it calls 'J-Space,' a term derived from the mathematical Jacobian technique used to analyze its cognitive processes. The company highlighted that Claude can engage with concepts in J-Space while performing other tasks, such as copying sentences. Anthropic's research paper frequently mentions 'consciousness' but stops short of asserting that its models are conscious. The company also indicated that monitoring J-Space could help identify misalignment or problematic behavior in AI models.