Carbon monitoring is becoming ever more critical as climate change accelerates and society turns to carbon management strategies, ranging from carbon credits to the preservation and restoration of natural carbon sinks. Yet the success of these approaches depends on robust science: measurements must be accurate, verification must be rigorous, and promises must be grounded in evidence. Machine learning (ML) is rapidly transforming carbon cycle research, offering new opportunities to integrate diverse data streams, harness remote sensing, and connect multiple lines of evidence across scales. This session will highlight recent advances in ML applications for investigating, monitoring, and managing the carbon cycle, spanning satellite-based greenhouse gas estimation, biomass and forest monitoring, soil and peatland carbon dynamics, wetland and ecosystem restoration, and the mapping of terrestrial and oceanic carbon storage. We particularly encourage contributions that address hybrid modeling, uncertainty quantification, ecological mapping, knowledge-guided and trustworthy ML in carbon markets and policy contexts. By bringing together advances from Earth observation, process modeling, and policy-relevant applications, this session aims to explore both the promises and challenges of ML in delivering actionable insights for carbon management and climate mitigation.