1.3× to 2.9× Energy Boost Signals Quantum Algorithm Backdoor

Researchers demonstrated a new vulnerability in variational quantum algorithms (VQAs) triggered by a standard optimization technique, revealing a potential supply-chain security risk. The team, including Ahatesham Bhuiyan and Mengxin Zheng, discovered the “CutBackdoor” attack, which exploits circuit cutting, a necessary process when quantum circuits exceed hardware capacity, to amplify energy levels in specific computational paths. Experiments on IBM quantum backends showed the attack boosted energy in cut paths by a factor of 1.3 to 2.9 without modifying the circuit itself. Crucially, no attacker presence at deployment is needed, as the trigger relies solely on the victim invoking circuit cutting due to qubit limitations, a unique characteristic differentiating it from prior quantum backdoor attempts.

Variational Quantum Algorithms (VQAs) face a newly discovered vulnerability stemming from a standard optimization technique; researchers have demonstrated a backdoor attack activated by circuit cutting, a process essential for running complex algorithms on limited quantum hardware. The work, led by Ahatesham Bhuiyan and Mengxin Zheng, details how maliciously crafted parameters, distributed through public repositories, can remain undetected until a victim invokes the CutQC workflow to address qubit capacity issues. The “CutBackdoor” attack exploits the fact that circuit cutting, while necessary, introduces a unique attack surface; poisoned parameters subtly increase reconstruction error along the cut paths without impacting full-circuit validation performance. Experiments conducted on IBM quantum backends revealed a significant energy boost in these cut paths, ranging from 1.3 to 2.9 over clean baselines on VQE and VQD benchmarks, while maintaining a low stealthiness error on the full circuit.

The researchers found the persistence of this effect across different backends and cut placements, even with matched compilation, suggests a structural weakness in VQA security. While Zero-Noise Extrapolation offered partial mitigation, the attack remained effective, and the diagonal-cost QAOA benchmark helped define the attack’s limits; the team’s theoretical analysis and empirical validation confirm the potential for supply-chain vulnerabilities in VQA deployments.

Variational Quantum Algorithms (VQAs) increasingly rely on pre-trained parameters sourced from public repositories, creating a potential vulnerability in the quantum computing supply chain that has largely been overlooked. This attack specifically targets the CutQC circuit cutting workflow, highlighting a previously unconsidered attack surface within a common and essential optimization technique for near-term quantum computing.