AI-driven carbon and green accounting: architectures, applications, and the AI sustainability paradox

Abstract

Carbon accounting and green accounting have converged with artificial intelligence over the past five years, yet the resulting literature remains split across research communities that rarely cite one another. This study adopts a structured critical review rather than an exhaustive systematic review. A curated corpus of fifteen studies published between 2020 and 2026 is examined to identify recurring technological pathways, methodological patterns, and unresolved accounting issues at the intersection of AI, carbon accounting, and green accounting. Three technical pathways emerge. Data-centric systems fuse enterprise resource planning records, sensor streams, and machine-learning emission prediction. Ontology-driven hybrids target small and medium enterprises through semantic gap-filling under sparse data. Real-time tagging platforms combine blockchain traceability, digital twins, and behavioural personalisation to push product-level carbon information to consumers. Empirical correlations between green accounting practice and economic outcomes are documented across multiple country contexts, alongside self-reported carbon footprint reductions of up to sixty percent that carry very-low certainty. A cross-cutting tension runs through the corpus: AI simultaneously enables finer-grained carbon measurement and draws on emissions-intensive training and inference workloads. The review responds with the Net Sustainability Intelligence framework, an analytical lens that formalises the relationship between the carbon insight produced by an AI-enabled accounting system and the carbon cost of producing that insight, and sketches three architectural extensions whose operational details remain open for future empirical work.