Harvest sustainability is a primary goal of wildlife management and conservation, and in a changing world it is increasingly important to consider environmental drivers of population dynamics alongside harvest in cohesive management plans. This is particularly pertinent for harvested species that are acutely experiencing effects of climate change. The Pacific walrus (Odobenus rosmarus divergens), a critical traditional subsistence resource for indigenous communities, is simultaneously subject to rapid habitat loss associated with diminishing sea ice and an increasing anthropogenic footprint in the Arctic. We developed a theta-logistic population modeling-management framework to evaluate various harvest scenarios combined with four potential climate/disturbance scenarios (ranging from optimistic–pessimistic) which simulates Pacific walrus population dynamics to the end of the 21^st century. We considered two types of harvest strategies: (1) adaptive harvest scenarios wherein harvest is calculated as a percentage of the population and annual harvests are updated at set intervals as the population is reassessed, and (2) non-adaptive harvest scenarios wherein annual harvest remains constant. All climate/disturbance scenarios indicated declines of varying severity in Pacific walrus abundance to the end of the 21^st century, even in the absence of harvest. However, we found that an adaptive annual harvest of 1.23% of the independent-aged female subset of the population (e.g., 1,280 independent-aged females harvested in 2020, representing contemporary harvest levels) met our criterion for sustainability (>70% probability of maintaining population abundance above maximum net productivity level) under all climate/disturbance scenarios, accepting a medium risk tolerance level of 25%. This suggests that the present rate of Pacific walrus harvest is sustainable and will continue to be—provided the harvest adapts to match changes in population dynamics. Our simulations suggest that a sustainable non-adaptive harvest is also possible, but only at low levels if the population declines as expected. Applying a constant annual harvest of 1,280 independent-aged females (equivalent to contemporary harvest levels of 1.23) exceeded our criterion for sustainability and resulted in a >5% chance of quasi-extinction by the end of the 21^st century under three of the four climate/disturbance scenarios we evaluated. Our results highlight the importance of adaptive co-management strategies, and we suggest such modeling frameworks are useful for managing for harvest sustainability in a changing climate.