<![CDATA[Artificial intelligence (AI) systems are now used widely across important fields, raising concerns about long-term risks, accountability, and how these systems are managed after launch. Most AI governance frameworks emphasize development, deployment, and monitoring; however, they typically lack explicit procedures for retiring or decommissioning AI systems. The absence of this provision complicates efforts to address regulatory noncompliance, ethical violations, inadequate performance, or ongoing harm that cannot be resolved through minor adjustments. This paper conducts a literature review to examine how the decommissioning of AI systems is addressed from a governance perspective. A concept-driven methodology identified seven peer-reviewed studies that examined instances where AI systems were suspended, withdrawn, or replaced following their implementation. The findings indicate that decommissioning is primarily seen as a strategy for managing unresolved post-deployment risks, rather than just a technical shutdown. Decommissioning practices are fragmented across multiple domains and lack full integration within the AI lifecycle. This paper introduces a governance-oriented framework for decommissioning artificial intelligence systems, building on the preceding review. The framework delineates the appropriate initiation of decommissioning, assigns decision-making responsibilities, specifies required actions at each system level, and outlines post-decommissioning procedures. By conceptualizing decommissioning as a distinct phase within the AI lifecycle, the framework clarifies governance structures and highlights the need for further research and improved system-retirement practices.]]>
Copyrights © 2026
