Approximately 240 million children, 10% of children worldwide, have disabilities. However, few studies have investigated requirements for AI-based tutoring and assessment for this population within a real-world context. It is unclear how special education classrooms experience these emergent tools in practice. We therefore conducted interviews with 18 special education teachers in the United States about the usage, challenges, and perceived benefits of AI in their classrooms. The interviews reveal tensions between special education’s need for personalized learning systems and difficulty integrating existing techno-solutions due to: (1) poor interface and curriculum adaptations for students with learning disabilities, (2) under-consideration of special education’s additional form of academic assessment when differentiating students, and (3) criteria for subject mastery incompatible with special education students’ personalized learning plans. Nevertheless, special education teachers remain optimistic towards AI’s promise of flexible instruction for students requiring accessible digital interfaces. From these findings, we conclude with design implications and potential research directions to better synergize emerging techno-solutions with special education classrooms’ needs. This work envisions tangible pathways towards inclusive AI for students with disabilities and a more equitable classroom of tomorrow.