Global environmental change has motivated multiple interventions in pursuit of sustainable outcomes within tropical forest landscapes. Fire is recognised as a key stressor facing forest conservation efforts. Large-scale accidental fires are increasingly prevalent across the forested tropics, generating negative impacts across sectors and scales. Policy responses to mega-fires in the Brazilian Amazon have been diverse but all are dominated by an anti-fire narrative that highlights long-stigmatised smallholder agricultural practices. Despite forest conservation initiatives and fire management policies, escaped fire (wildfire) remains pervasive. Forest conservation initiatives are often situated in contexts where swidden agriculture prevails, generating a need for an improved understanding of the interplay between fire management and conservation initiatives on the ground. We explore these dynamics through a case study approach in three leading forest conservation initiative types, situated across diverse contexts in the Brazilian Amazon: a Reduction of Emissions of Deforestation and Degradation (REDD+) site (in Middle Solimões region), an extractive reserve (RESEX) (in Arapíuns region), and a Green Municipality Pact (GMP) (in Paragominas). Between sites, climate and colonisation histories vary, yet all demonstrate that farmers experience the burden of escaped fire, attesting to the failure of fire management policies and suggesting that fire (as currently managed) threatens forest conservation goals. Restrictive fire management policies do not replace the necessity of fire-based agriculture and rather serve to disempower swidden farmers by making burning increasingly illicit. We show that awareness of fire-free alternatives exists, yet experience is limited and constraints are considerable. We argue that marginalising fire use in the context of forest conservation initiatives contributes to a legacy of failed interventions and jeopardises partnerships between communities and conservation practitioners. Finally, we suggest that given the absence of imminent and viable fire-free alternatives, particularly in sites where swidden and conservation collide, a new model of fire warrants experimentation.