Large, multi-fault earthquakes increase the threat of strong ground shaking and reshape the probability of future events across a system of faults. Fault junctions act as conditional barriers, or earthquake gates, that stop most earthquakes but permit junction-spanning events when stress conditions are favorable. Constraining the physical conditions that favor multi-fault earthquakes requires information on the frequency of isolated events versus events that activate faults through the junction. Measuring this frequency is challenging because dating uncertainties limit correlation of paleoseismic events at different faults, requiring a direct approach to measuring rupture through an earthquake gate. We show through documentation and finite-element modeling of secondary fault slip that co-rupture of the San Andreas and San Jacinto faults (California, USA) through the Cajon Pass earthquake gate occurred at least three times in the past 2000 yr, most recently in the historic 1812 CE earthquake. Our models show that gate-breaching events taper steeply and halt abruptly as they transfer slip between faults. Comparison to independent chronologies shows that 20%–23% of earthquakes on the San Andreas and the San Jacinto faults are co-ruptures through Cajon Pass.