Anthropic announced Monday that it has discovered a small internal workspace within Claude, its AI model, where the system holds and manipulates ideas without converting them into words. The company likens this structure to how humans consciously access thoughts, though it has not demonstrated that Claude experiences feelings or sensations.
The finding provides fresh ammunition for the ongoing debate over machine consciousness, revealing a division between deliberate reasoning and automatic computation. This human-like separation suggests Claude can engage in silent deliberation, separate from the “chain of thought” reasoning it displays to users.
In a video and accompanying post on X, Anthropic stated, “We can see Claude silently perform reasoning steps in its head—noticing bugs in code, identifying images, and more.” The company has named this phenomenon “J-Space,” after the Jacobian mathematical technique used to detect it.
Anthropic described Claude’s ability to “activate concepts similar to how humans can think about one thing while doing another.” The finding underscores a growing ability to peer into AI's inner workings, raising questions about transparency and accountability in model behavior.
However, critics caution that anthropomorphizing model internals risks overstating what is ultimately statistical pattern matching. The company has not demonstrated subjective experience, leaving the line between sophisticated computation and consciousness unresolved.