Anthropic Reveals Human-Brain-Like Cognitive Structure That Emerged Spontaneously in Claude

Neuroscience’s global workspace theory holds that the brain sends a small amount of important information into a shared channel that different systems can access. Anthropic found a functionally similar structure in Claude, called J-space, that emerged independently during training. The finding suggests complex cognition may converge on similar architectures. However, the research covers only reportable, actionable information processing and does not mean Claude has feelings or subjective consciousness.

Anthropic published the research on July 6, 2026, using a Jacobian lens to examine Claude’s internal representations. J-space holds only dozens of concepts at a time and accounts for less than 10% of overall internal activity, yet some regions have read-and-write connections to network components that are about 100 times stronger than those of ordinary representations. Removing J-space had little effect on fluent conversation, but reduced multistep reasoning performance to nearly zero. The research did not involve any transaction or investment amounts.