Globally, improved forest management (IFM) practices are recognized as powerful, low-cost natural climate solutions, but carbon leakage may reduce project efficacy, and uncertainty about the impact of leakage may result in underinvestment in improved forest management. While some carbon project protocols deduct predicted leakage impacts before issuing credits, the methods and rates applied are extrapolated from a small amount of evidence. This paper applies a data-driven approach to quantify project-level forest leakage impacts at regional and global scales. We use a dynamic global forest sector model to estimate the leakage effects of two forest carbon project interventions under varying implementation rates and conditions: extended rotations and permanent set asides. We then apply statistical methods to identify key drivers of varying leakage estimates. We find that leakage is considerably lower than expected—carbon leakage rarely passes 50% and is often negative—especially in the tropics and for set asides. However, rates vary considerably by policy design and are influenced by the project type, measurement period, project implementation rate, and dynamic market and ecological response to harvest schedule changes. The regionalized, condition-specific leakage estimates from this paper provide the most detailed global assessment of carbon project leakage yet assembled. These data provide an evidence base from which to discount forest management carbon projects where needed and can help ensure accurate accounting of IFM interventions’ net climate benefit.