AI offers unprecedented opportunities for climate adaptation, hazard response, and environmental governance, yet we are not thinking enough about the risks when these tools fail under the nonstationary, extreme, and compounding conditions they were designed to address. This session examines the coupling between accelerating AI deployment and escalating Earth-system stress across three timescales. Acute (hours–days): During compound hazards, AI-driven forecasts and emergency systems become critical single points of failure. What are the consequences of AI breakdown during out-of-distribution events? Institutional (seasons–decades): Agencies depend on AI for water allocation, insurance pricing, and infrastructure planning. How does automation bias interact with climate nonstationarity? Governance (decades–century): Autonomous monitoring and AI-optimized grids raise accountability questions when algorithms operate beyond human deliberation speed. By confronting these risks, the community can build resilience to realize AI's transformative potential. We welcome contributions spanning AI failure modes, compound risk, resilience, and governance of autonomous environmental systems.