New Research Defines 'Context Access Divide' as Hidden AI Inequality

A new 19-page paper by researcher Masahiro Fujita (arXiv:2607.08495) introduces the "Context Access Divide" (CAD), a hidden dimension of AI inequality that operates at the individual interaction level. Prior work by Sharp et al. (2025) defined agentic inequality across three dimensions—availability, quality, and quantity of AI agents—but Fujita argues these miss a critical factor: how the AI system accesses a user's accumulated knowledge. Two users with the same agent may experience dramatically different utility depending on whether the system autonomously retrieves context from the user's knowledge corpus (Dynamic Context Retrieval) or requires the user to manually attach documents at each query (Manual Attachment). For knowledge workers with tens of thousands of files, the burden of manual context curation reproduces the inefficiencies AI was supposed to eliminate, creating a qualitative threshold in usefulness.

Fujita formalizes this divide with a probabilistic model grounded in the fan effect literature from cognitive psychology. The model demonstrates that manual context attachment leads to a combinatorial collapse in task-success probability as corpus size and task conjunctivity grow—meaning that the more documents a user has, the harder it becomes to manually attach the right ones. Dynamic retrieval architectures (based on MCP and RAG) are structurally insulated from this collapse because they can autonomously surface relevant context. The paper frames "contextuality" as a new dimension of AI-mediated inequality and discusses its implications for knowledge-work stratification and AI platform governance. It calls for attention to interaction-level design choices that could either widen or narrow the divide.