Amid global disparities in technological advancement and carbon emissions, this study evaluates the role of artificial intelligence (AI) technology innovation in improving carbon emission efficiency. AI innovation is assessed using country–year patent data, distinguishing between technology-oriented and application-oriented domains. To measure carbon emission efficiency while accounting for technological heterogeneity across countries, we develop a machine learning–based meta-frontier evaluation framework. This framework provides complementary assessments of efficiency from the perspectives of the meta-frontier, group frontiers, and the technology gap. Results reveal that AI technology innovation significantly improves carbon emission efficiency and narrows technological gaps. Technology-oriented AI exerts stronger effects than application-oriented AI, and the relationship between AI and efficiency follows an inverted U-shape, with the largest gains observed in middle-tier technology groups. Rising income levels further strengthen both the magnitude and persistence of these impacts. Mechanism analysis shows that AI enhances efficiency primarily through technological progress, while regulatory quality and clean energy adoption serve as enabling conditions, and market forces alone remain insufficient. These findings demonstrate that AI can reduce global carbon inequalities, but its sustainability potential depends critically on supportive governance and clean energy transitions. © 2025 ERP Environment and John Wiley & Sons Ltd.