From Ethical AI to Sustainable AI: Why Environmental Materiality Belongs in Every AI Risk Assessment

The rapid expansion of artificial intelligence (AI) introduces significant environmental sustainability challenges across electricity consumption, greenhouse-gas emissions, freshwater use, semiconductor manufacturing and infrastructure lifecycles. Although existing AI governance standards increasingly acknowledge environmental impacts, practical mechanisms for integrating environmental evidence into enterprise AI decisions remain fragmented. This paper proposes the 4M Environmental AI Governance Framework, comprising Materiality, Measurement, Mitigation and Monitoring, with Accountability and Disclosure embedded throughout the AI lifecycle. The framework integrates environmental assessments across three interconnected levels: product and project, organisation, and ecosystem. It introduces a materiality-screening methodology, decomposed environmental metrics, impact-hotspot identification, sustainability trade-off assessment and environmental lock-in considerations. It also proposes a cross-provider approach for extracting AI workload information and connecting operational telemetry with environmental estimates. An illustrative enterprise case study compares small language model, large language model and agentic AI architectures to demonstrate how environmental evidence may inform risk committee decisions. The framework is presented as a conceptual and practitioner-oriented methodology, rather than an empirically validated standard. It aims to support transparent, proportionate and continuously governed environmental decision-making in enterprise AI systems.