Artificial intelligence is now embedded across the immigration system. IRCC has published an Algorithmic Impact Assessment for its Study Permit Extension Eligibility Model, confirming that automated decision-making is already operational in Canadian immigration. Settlement organizations are adopting AI chatbots and triage tools under funding pressure, and newcomers increasingly rely on commercial AI systems for information and guidance. In the May 2026 non-academic partner consultation, partners repeatedly raised concerns about discrimination based on language, culture, or background, opaque automated screening, and the unknown implications of AI-driven decision-making in immigration processes. Existing accountability tools treat these systems as black boxes: they test inputs and outputs but cannot explain why a system behaves as it does, and they can miss bias that is hidden rather than removed. Recent advances in AI interpretability make it possible to examine the internal concepts a model has learned, including concepts about nationality, religion, immigration status, and emotional expression, and to trace how those concepts shape the model's outputs. This project applies these methods to AI systems in immigration contexts for the first time. It will produce evidence on whether and how such systems encode bias, methods for auditing the closed commercial systems that governments and platforms actually deploy, and practical tools and capacity for the settlement sector. Within Cluster 3, it complements existing work on the adoption, politics, and ethics of advanced digital technologies by adding the capability to examine what these systems do internally.