Artificial intelligence (AI) is increasingly present in daily life and institutional decisions, yet its rapid development has outpaced ethical frameworks that address cognitive diversity. Neurodivergent individuals-including those with autism, ADHD, dyslexia, and Tourette syndrome-are systematically excluded by AI systems designed around neuronormative assumptions. This article examines how current AI technologies, from automated hiring to customer service and educational platforms, reproduce and amplify structural discrimination by failing to accommodate diverse cognitive profiles. A critical review of the European legal and policy framework as of 2025 (including the GDPR, AI Act, and Digital Rights Declaration) reveals that neurodivergence and cognitive accessibility remain largely unaddressed. In response, we propose a model for inclusive AI governance based on five pillars: participatory co-design, ethical oversight committees, dynamic regulation, ethical education in cognitive diversity, and representative data. We argue that only by integrating neurodivergence as a structural axis in design, regulation, and evaluation can AI systems become truly fair and just, ensuring technological progress benefits all members of society and preventing the perpetuation of algorithmic injustice.