Reddit administrators have generally struggled to prevent or contain such\ndiscourse for several reasons including: (1) the inability for a handful of\nhuman administrators to track and react to millions of posts and comments per\nday and (2) fear of backlash as a consequence of administrative decisions to\nban or quarantine hateful communities. Consequently, as shown in our background\nresearch, administrative actions (community bans and quarantines) are often\ntaken in reaction to media pressure following offensive discourse within a\ncommunity spilling into the real world with serious consequences. In this\npaper, we investigate the feasibility of proactive moderation on Reddit --\ni.e., proactively identifying communities at risk of committing offenses that\npreviously resulted in bans for other communities. Proactive moderation\nstrategies show promise for two reasons: (1) they have potential to narrow down\nthe communities that administrators need to monitor for hateful content and (2)\nthey give administrators a scientific rationale to back their administrative\ndecisions and interventions. Our work shows that communities are constantly\nevolving in their user base and topics of discourse and that evolution into\nhateful or dangerous (i.e., considered bannable by Reddit administrators)\ncommunities can often be predicted months ahead of time. This makes proactive\nmoderation feasible. Further, we leverage explainable machine learning to help\nidentify the strongest predictors of evolution into dangerous communities. This\nprovides administrators with insights into the characteristics of communities\nat risk becoming dangerous or hateful. Finally, we investigate, at scale, the\nimpact of participation in hateful and dangerous subreddits and the\neffectiveness of community bans and quarantines on the behavior of members of\nthese communities.\n