Anthropic researchers have published a comprehensive analysis identifying a distinct internal computational workspace within Claude, designated as J space. Detected using a mathematical instrument termed the Jacobi lens, this subnetwork occupies less than ten percent of the model’s total activation. It functions as a transient channel for reasoning steps, latent judgments, and multi-step deductions that frequently never appear in the final output. The workspace demonstrates three core properties aligned with Stanislas Dehaene and Lionel Naccache’s Global Neuronal Workspace Theory of human cognition: reportability, limited capacity, and flexible downstream integration. Anthropic’s experiments confirm that J space is an active processing hub rather than a passive log. Direct interventions successfully reroute the model’s internal processing; replacing a conceptual token for football with rugby alters the final answer, while swapping France for China consistently updates factual responses across multiple independent tasks. The space currently supports roughly twenty-five concurrent concepts. Independent verification by DeepMind researchers across multiple model architectures reinforces the validity of these findings. Experts note critical architectural divergences from biology, observing that J space emerges through feedforward neural passes instead of recurrent cortical loops, and exceeds human working memory limits. Analysts emphasize that while the evidence strongly supports machine access consciousness, it remains functionally distinct from phenomenal consciousness or subjective experience. Anthropic acknowledges current technical constraints, including the Jacobi lens’s reliance on single-token approximations and an unresolved mechanism governing workspace entry. Nevertheless, the research delivers the first empirical map of a manipulable, functional internal workspace in large language models. The discovery advances AI interpretability, provides novel levers for steering model reasoning, and establishes a quantitative bridge between artificial architectures and computational cognitive science.