Anthropic published a research paper Sunday revealing that its Claude language models have spontaneously developed an internal structure that mirrors one of the most influential theories of how human consciousness works. The finding, the company says, has already begun reshaping how it monitors its AI systems for safety risks.
The 16-author study, titled "Verbalizable Representations Form a Global Workspace in Language Models," describes how researchers used a new mathematical technique to peer inside Claude. They discovered what they call a "J-space" — a small, privileged zone of internal activity where the model holds concepts it can report on, reason with, and direct at will, surrounded by a larger ocean of automatic processing it cannot access.
The researchers present evidence that "an analogous functional distinction has emerged in modern AI models" to what exists in humans. This mirrors the global workspace theory of consciousness, which proposes the brain has a central hub for conscious thought surrounded by unconscious processing.
The work lands amid an intensifying scientific debate over whether machines can possess anything resembling a mind. While the paper does not claim Claude is conscious, it shows an AI system converging on a functional architecture analogous to a key component of human cognition.
Anthropic says the discovery is already being applied to safety monitoring, potentially offering new ways to detect when models are operating on unverifiable internal states. However, critics may argue that similarity in function does not imply similarity in subjective experience.