It is well recognized that military alliances can provide their members with important security benefits. However, less attention has been paid to the policy concessions states must grant others to enter into military alliances. To study this aspect of alliances, I develop a three-actor bargaining model of alliance formation that endogenizes both external threat and policy concessions. Specifically, a target state bargains with a potential defender over the concessions it must make to ally and then responds to a potential challenger. The model suggests that what is important for policy concessions in alliances is not just the power of the threatened state but its power relative to its challenger and how an alliance will change the distribution of power. I test implications of this model using data on promised policy concessions formalized in alliance treaties and find strong support for the hypotheses. More specifically, I find that states are willing to make more concessions in exchange for an alliance when they are unlikely to defeat their challengers alone and when their allies have a large effect on their probability of winning in war. These findings refine existing theories of alliances and offer the first large-N analysis of policy concessions in alliances.