Artificial Intelligence in Climate Policy Governance: Evidence Boundaries, Sustainability Risks and Institutional Safeguards

Artificial intelligence (AI) supplies increasingly sophisticated information for climate policy, yet its decision value depends on how institutions use that information. This structured critical review compares 121 sources across policy analysis, emissions monitoring, targeting, carbon markets and climate finance. It examines study designs, comparators and institutional settings to explain the transition from information production to action and outcomes. Validation is task-specific: document coverage, emission detection and policy interpretation require different benchmarks. Targeting gains depend on the capacity to reallocate resources, while more detailed monitoring leaves counterfactual, eligibility and accounting judgements unresolved. Few sources document AI use in a climate policy workflow, and observed policy outcomes cannot be inferred from technical performance alone. The review connects these findings to four governance tests: credible information, feasible action, authorised and reviewable use, and whole-workflow accountability. The resulting conceptual framework provides a guide to AI adoption and evaluation, with its applicability across institutional settings awaiting empirical validation.