Global Workspace is a cognitive architecture concept, adapted from Global Workspace Theory in consciousness studies, describing a narrow, shared broadcast channel through which information from specialized subsystems is selected, integrated, and made widely available to the rest of a system. In neural network research the term has been repurposed to describe an analogous bottleneck that models appear to develop internally — a compressed, relatively interpretable representation that gets routed to many downstream components, in contrast to the diffuse, entangled computation found elsewhere in the network. Its emergence is treated as evidence that large models may organize information processing around a small number of legible communication channels rather than uniformly distributed representations.