CrossEcho: Predictive Modelling of Crosstalk Side‐Channel Leakage on Multitenant Superconducting QPUs

Multitenant scheduling on cloud‐based quantum processors introduces crosstalk—unwanted interqubit interactions that can be exploited as physical side‐channels. While prior work has demonstrated that crosstalk leaks victim circuit information with high classification accuracy, all existing characterisations are post hoc: They observe, fingerprint or reconstruct the induced error after the attack executes. We present CrossEcho—named for the ZZ phase ‘echo’ of a victim’s gate operations imprinted on colocated attacker qubits through the coupling fabric (referring specifically to phase accumulation, not population transfer or spin‐echo physics)—a forward predictive model that forecasts crosstalk leakage magnitude from device calibration data, qubit placement and circuit properties before measurement. The model derives effective ZZ coupling strengths from transmon perturbation theory, propagates calibration uncertainty via Monte Carlo sampling ( n = 1000) and is evaluated using a dual hold‐out protocol (33% of qubit placements held out) on 46 cells across two IBM Quantum Heron r2 processors. A dedicated readout‐only control experiment—the victim circuit has zero gates—isolates measurement‐induced crosstalk from gate‐induced ZZ coupling via their qualitatively different distance‐decay signatures. The gate‐crosstalk excess over the readout baseline is small but statistically significant ( p = 0.03, Cohen’s d = 0.78), requiring aggregate analysis for reliable detection. Cross‐processor‐architecture validation on the Origin Quantum Cloud WK_C180_2 processor (2D‐grid topology) versus IBM Heron r2 (heavy‐hexagonal topology) reveals an order‐of‐magnitude increase in crosstalk magnitude on the denser 2D grid, directly confirming the architecture dependence that the perturbation‐theory model identifies as its dominant uncertainty source. We establish the readout‐only control as essential methodology for multitenant crosstalk studies, quantify the shot‐count and circuit‐depth thresholds needed to isolate the ZZ signal, and propose coarse frequency binning by cloud providers as a practical path to deployable forward crosstalk risk assessment.