Anthropic's new "J-lens" reveals a silent workspace inside Claude that mirrors a leading theory of consciousness

Anthropic published research describing a new interpretability method, the J-lens, that identifies a small internal zone in Claude models where reportable and controllable concepts appear to reside. The company says this J-space resembles global workspace theory, in which a narrow set of information becomes available for flexible reasoning while most processing remains automatic and inaccessible.

The findings do not prove AI consciousness, but they may intensify debate over whether advanced models are developing mind-like functional structures. More practically, Anthropic’s work points to interpretability becoming central to AI safety, giving researchers new ways to monitor hidden reasoning and potential risk signals inside frontier models.