We used incentivized experimental games to manipulate leader power — the number of followers and the discretion leaders had to enforce their will. Leaders had complete autonomy in deciding payouts to themselves and their followers. Although leaders could make prosocial decisions to benefit the public good they could also abuse their power by invoking antisocial decisions, which reduced the total payouts to the group but increased the leaders' earnings. In Study 1 (N = 478), we found that both amount of followers and discretionary choices independently predicted leader corruption. In Study 2 (N = 240), we examined how power and individual differences (e.g., personality, hormones) affected leader corruption over time; power interacted with endogenous testosterone in predicting corruption, which was highest when leader power and baseline testosterone were both high. Honesty predicted initial level of leader antisocial decisions; however, honesty did not shield leaders from the corruptive effect of power.