Artificial intelligence (AI) is reshaping higher education and raising urgent questions about whether emerging technologies will advance or undermine inclusive participation. This narrative review synthesizes a purposive body of evidence published between 2020 and 2026 on the intersection of AI and inclusion in higher education. It examines three interconnected dimensions: AI-powered assistive technologies for students with disabilities; AI-driven personalization and its alignment with Universal Design for Learning (UDL); and the risks of algorithmic bias for students who have historically been marginalized in educational systems. The review draws on peer-reviewed studies, systematic and scoping reviews, policy documents, and Brazil-based studies focusing on deaf inclusion and Brazilian Sign Language (Libras) technologies. Findings indicate that AI can reduce barriers for students with physical, sensory, cognitive, and learning disabilities when it is implemented through accessible design, institutional support, and participatory governance. However, the evidence also shows that AI systems trained on non-representative data can replicate or amplify inequities, particularly when disability, linguistic diversity, race, income, and digital access are treated as secondary concerns. The article concludes that inclusive AI in higher education requires a rights-based approach grounded in UDL, co-design with affected communities, transparent governance, and continuous evaluation of access, bias, privacy, and learning outcomes.