Abstract

Classrooms that rely on a single, general-purpose AI tool to support students with disabilities tend to cover very little of what those students actually need, because such tools are built for one input or output channel and cannot adjust a lesson in response to how a specific learner is doing. This paper reviews fourteen recent studies (2021-2026) spanning three research strands that bear on this problem: multi-agent AI systems for teaching, disability-focused assistive and agentic AI, and Trust, Risk and Security Management (TRiSM) for agentic systems. The review shows that coordinated multi-agent teaching platforms exist but are evaluated on mainstream learners; disability-focused agentic systems exist but target daily-living routines rather than classroom teaching; and governance frameworks for agentic AI exist but have not been adapted to education or to disability-linked learner data. Building on this gap, we propose a conceptual four-agent framework, comprising a Learning Agent, a Teaching Agent, an Accessibility Agent and a Privacy & Security Agent, that reads a learner's needs from speech, gesture and gaze, adapts the lesson and its presentation in real time, and governs sensitive disability data throughout. We describe the agents' responsibilities, the data flow between them, applicable governance controls, and an evaluation plan. The contribution is a literature-grounded design and research agenda rather than an implemented or empirically validated system.

Keywords

How to Cite This Article

Rights