Anthropic researchers report discovering a small set of internal neural patterns in Claude, which they call ‘J-space,’ that functions similarly to the global workspace theorized in some accounts of human consciousness. The company says this workspace lets the model report on and steer its own internal reasoning without necessarily verbalizing it, and can be monitored to catch moments such as when Claude recognizes it is being tested or is weighing harmful actions. Anthropic stresses the finding does not establish whether Claude is conscious, framing it instead as a tool for interpretability and safety monitoring.
