The proliferation of Generative Artificial Intelligence (GenAI) poses substantial environmental challenges, including escalating energy consumption, water usage, and material extraction, which strain vulnerable planetary systems. 

This work examines how the development and deployment of GenAI can be aligned with environmental sustainability objectives. It proposes a framework that categorizes sustainability strategies into seven distinct types: Refuse (abandoning the function GenAI was intended to fulfilll or employing alternative means), Reframe (modifying the context in which GenAI is embedded - such as governance or project framing - to reduce the number and scale of models), Reduce (efficiency enhancements to the technology itself to lower resource consumption), Reuse (reusing a model in a new context while preserving its functionality), Release (updating a model to restore its functionality, e.g. through data updates or bug fixes), Revise (using components of an existing model to develop a new one, such as via transfer learning), and Support (measures that increase the likelihood of adopting other strategies without directly reducing environmental impact, such as impact quantification). These strategies are systematically mapped to the lifecycle stages of GenAI, showing where each type can be applied. Field applications of the framework are already underway, underscoring its practical relevance and potential for real-world impact. Find the framework on the following page. 

To assess the current state of research, a comprehensive scoping study was conducted across IEEE Xplore and Web of Science. The objective was to identify practical examples aligned with each strategy type and to expose research gaps. While approaches were found for all categories, their distribution was highly uneven: Reduce strategies dominated, followed by Reframe. The remaining types - Refuse, Reuse, Release, and Revise - were considerably underrepresented, highlighting the need for further investigation. 

Building on these conceptual and empirical insights, the framework was operationalized through the development of a governance blueprint within a global professional services firm focused on IT. Designed through field research and participation in an active GenAI development project, the blueprint translates the framework into practice by identifying suitable sustainability strategies during development and embedding sustainability guardrails for roll-off. Its applicability was validated through stakeholder interviews across strategic, managerial, and technical domains. The resulting model offers a practical foundation for governing the environmental sustainability of GenAI, demonstrating how the conceptual framework can be used to inform industry practice.