<p>This study analyses the amplification and diffusion of climate misinformation on Twitter during the COP26 and COP27 climate conferences. Drawing on a dataset of over 12 million English-language tweets, we combine machine learning classification, social network analysis, and qualitative content analysis to map how misinformation circulates across user communities. Climate misinformation is understood as content that denies or undermines the scientific consensus on anthropogenic climate change. Using a machine learning classifier trained on annotated climate datasets, tweets were labelled and assigned misinformation probabilities. Using community detection, we were able to distinguish between misinformation, non-misinformation, and mixed communities. Our findings show that misinformation does not remain isolated within echo chambers; instead, it often flows outward, particularly into mixed communities, which serve as key intermediaries between misinformation and non-misinformation communities. Through centrality measures, we identified a small set of influential user accounts that function as amplifiers and brokers of misinformation, both intentionally and inadvertently. These key users exhibit varying patterns of visibility, engagement, and connectivity, where we found two prominent user types to be those of the broadcaster and the mediator. The results show that the dynamics of misinformation dissemination are shaped by both content virality and underlying network structures.</p>