The threshold at which density-limiting plasma instabilities occur in tokamak fusion reactors is not well understood. Here, we present the first large-scale empirical evidence of a common collisionality threshold present across multiple devices and confinement regimes. Machine-learning guided analysis of a diverse experimental database identifies a scaling in edge effective collisionality (${\nu }_{*}$) that bears resemblance to the widely-utilized Greenwald limit but yields a substantially lower false positive rate. This threshold decreases with dimensionless pressure ($\beta$), consistent with theoretical expectations for enhanced transport from electromagnetic effects on resistive ballooning mode turbulence. These results suggest that future burning plasma devices such as ITER and ARC could potentially achieve higher density, and thus fusion power, than previously expected.