The Signal
IBM Gate Patterns Bias Neighboring Qubits in New Tests
A Scientific Reports study finds that two-qubit gate sequences on IBM quantum processors can bias measurements of nearby spectator qubits. The paper reports 40,000-experiment results for selected setups and a strong association between circ
Quantum.world Team · Written with AI · fact-checked ·
In selected IBM processor tests, researchers ran 40,000 experiments to measure how two-qubit gate setups affected nearby qubits with no direct interaction. The results, published October 21 in Scientific Reports, show that circuit configuration can bias measurement outcomes on these “spectator” qubits. The authors define a successful effect as a bias that shifts the target’s measured state in either direction in more than two-thirds of the experiments.
The study examines superconducting quantum processors accessed through IBM’s commercial systems. Its central finding is not that every nearby qubit flips, but that sequences of gates can produce measurable changes beyond the control baselines. The size of the effect depends on how the gates are arranged.
Gate arrangements change spectator outcomes
“Using commercially available IBM quantum devices, we observe that sequences of two-qubit gates are associated with a measurable bias in the neighbouring spectator qubits without direct interaction, producing quantifiable changes in neighbouring spectator-qubit measurement outcomes that significantly exceed the corresponding control baselines,” write Fernando Almaguer-Angeles, Pedro Ruas-Dieguez, Akshata Shenoy Hejmadi and Marcin Pawłowski of the University of Gdańsk.
The paper compares different configurations of CNOT, or CX, gates and examines single-qubit operations as well. Its appendices present results associated with the IBM systems Kyiv, Brisbane and Osaka. For the Brisbane results, the researchers report a Cramér’s V value of 0.7, which they describe as a strong association between gate configuration and the measured spectator outcomes.
What the 40,000 experiments show
The 40,000-experiment figure appears in the paper’s captions for individual Brisbane and Kyiv results; it should not be read as one total sample covering every device and test. The authors also provide two-sided 95% Wilson confidence intervals for those plotted success rates. The threshold they use—more than two-thirds of trials showing a biased outcome—sets a specific criterion for counting a configuration as successful.
That distinction matters because crosstalk is not a single fixed error rate. The reported results tie the observed bias to gate choices and circuit arrangement, while the differing device results show that the effect varies across processors. The authors’ statistical analysis summarizes the relationship between configuration and measurement outcome; it does not establish that all circuits or neighboring qubits behave alike.
A hardware result, not a live cloud attack
The paper describes the effect as “Rowhammer-like”: repeated operations on one part of a system can influence a nearby target, by analogy with disturbance effects in classical memory. Here the relevant physical mechanism is crosstalk between qubits during gate operations. The authors say the IBM systems served as the technical platform for their investigation and state that the work is not a criticism of the company’s products or services.
The experimental evidence supports a narrower claim than a demonstrated attack on another customer’s computation. The paper measures how selected gate sequences change spectator-qubit outcomes on IBM hardware; those measurements alone do not show that one cloud user can alter another user’s simultaneous job. For processor users, the concrete finding is that the tested gate configurations produced outcome biases exceeding the control baselines, with some reported setups assessed across 40,000 experiments.