Autocratic cooperation is difficult to study. Democratic states usually disfavor autocratic cooperation partners because they are perceived as less reliable and do not sign agreements with them. While it is challenging to capture autocratic cooperation with traditional approaches such as signed alliance treaties, co-sponsorship at the United Nations General Assembly (UNGA) offers a valuable alternative. UNGA co-sponsorship is less binding than alliances, allowing states to cooperate more freely with one another. What is more, states are required to choose cooperation partners actively. This allows us to study how autocracies cooperate in the international system at a venue that overcomes common restrictions to autocratic cooperation. We construct co-sponsorship networks at the UNGA and use the Leiden algorithm to identify community clusters. Our multiclass random forest classification model supports our assumption and shows that regime type is associated with cooperation clusters in UNGA co-sponsorship networks.