Quantum Computers Can Now Keep Secrets

Quantum computers are increasingly accessed through the cloud, meaning users must trust third parties with their data. This poses a major barrier for sensitive applications like healthcare, where privacy is paramount. Researchers at Los Alamos National Laboratory have developed a homomorphic encryption technique for quantum annealing that allows computations on encrypted data without revealing the input or output to the cloud provider. This breakthrough could enable quantum computers to handle confidential information securely, closing a critical gap in their practical use.

The key finding is that homomorphic encryption for quantum annealing can be achieved with little or no performance cost. The researchers demonstrated this using spin reversal transformations, which modify the problem sent to the quantum computer in a way that hides the original data. After the computation, the user applies a secret key to decrypt the result. Unlike classical homomorphic encryption, which adds significant overhead, this maintains the quantum annealer's efficiency. For example, in tests on a nonnegative/binary matrix factorization problem, the average difference in performance between encrypted and unencrypted runs was just -0.6%, meaning encryption slightly improved in some cases.

Ology relies on spin reversal transformations, a concept from quantum physics. Essentially, the user generates a random secret key and uses it to transform the problem into an encrypted version before sending it to the quantum annealer. The annealer solves this encrypted problem, and the user decrypts the solution using the same key. This process ensures that the third party operating the quantum computer never sees the original data or the final answer. The approach is fully homomorphic, meaning it can handle any computation that a quantum annealer normally performs, from optimization to data analysis.

From three diverse problem types confirm 's effectiveness. In a hydrologic inverse problem—used to model aquifer properties—the average performance difference was -0.02%, again showing negligible impact. For RAN1 benchmark problems, which test random configurations, the difference was 6.1%, but the researchers attribute this to statistical variability rather than a systematic penalty. Figures 3, 4, and 5 in the paper illustrate the cumulative distribution functions of energies for these problems, showing that the encrypted and unencrypted versions produce nearly identical outcomes. This consistency across applications underscores the robustness of the technique.

This development matters because it addresses a fundamental limitation of cloud-based quantum computing. Currently, users cannot process sensitive data—such as health records under HIPAA or financial information—without trusting third parties. By enabling secure computations, this could accelerate quantum computing's adoption in fields like medicine, where data privacy is non-negotiable. Moreover, it narrows the performance gap with classical computing in scenarios requiring encryption, as classical s incur heavy computational costs for similar security.

Limitations include the need for further analysis on how the encrypted problem might reveal information about the secret key. The paper notes that for some problem classes, like RAN1, the transformation may not leak data, but rigorous security proofs are still needed. Future work will explore these aspects and extend to other quantum architectures, such as reverse annealing, to broaden its applicability.