Artificial intelligence (AI) holds significant potential for inclusive education through personalized learning, adaptive tools, and enhanced accessibility for diverse learners. At the same time, AI poses substantial risks to inclusion, including algorithmic bias, normative learner profiles, and unequal access that may deepen existing inequalities. How education policies frame these inclusive potentials and exclusion risks shapes whether AI adoption ultimately advances or undermines equitable participation in education. This paper presents a secondary analysis drawing on two interrelated datasets: (1)seven AI and education policy documents from Austria (n=4) and Switzerland (n=3), analyzed in a broader DACH study on inclusion and equity, and (2) 18 documents from all 16 German federal states, using the same criteria. Key questions are: Do these policies recognize AI's potential to foster inclusive education — through differentiation, individualization, and accessible learning environments? Do they also address the risks AI poses to inclusion, such as bias, digital divides, and discrimination? The analysis follows qualitative content analysis (Kuckartz & Rädiker, 2024) using a MAXQDA-based coding framework. The categories "Inclusive Potentials" (e.g., differentiation, accessibility) and "Exclusion Risks" (e.g., bias, technological barriers, discrimination) were specifically developed to capture how policies frame AI in relation to inclusive education. Preliminary findings indicate a clear imbalance: policy documents emphasize AI's potential for individualized learning and accessibility, while risks such as bias, digital inequality, and discrimination receive substantially less attention. Germany engages most comprehensively with both, whereas Austrian and Swiss documents highlight opportunities more prominently and address risks more selectively. The analysis of all German federal states examines whether this pattern persists across the decentralized education system. This paper argues that genuinely inclusive AI governance requires policy frameworks that balance AI's inclusive promise with critical awareness of its exclusion risks, otherwise, policies risk reinforcing a techno-solutionist narrative that obscures structural barriers to participation and equity.