A quantum cloud could hide your data and still prove its work

Imagine sending a valuable calculation to a remote quantum computer. You might want its operator to know nothing about your private input. You might also want somebody other than that operator to check the work.

Those wishes can sound contradictory: how do you check something you are not allowed to see? A paper highlighted by NTT Research as the FOCS computer-science conference opened on November 8 explores a mathematical route to combining those protections. It also addresses a separate problem: disguising a quantum program’s internal design while leaving it usable. These are theoretical results, not the launch of a protected computing service.^[1]

The conference provides a timely reason to explain the work, not a new discovery date. The arXiv record lists the paper’s first posting as October 9, 2025, and its latest submission-history entry as January 13, 2026. FOCS’s accepted-paper list independently confirms its inclusion. The paper is by James Bartusek, Aparna Gupte, Saachi Mutreja and Omri Shmueli.^[2][3]

Three secrets that should not be confused

Consider an illustrative company assessing candidate materials. Its input might contain a confidential design. Its software might encode a valuable method. Its eventual answer could influence which experiment it pays for next.

That creates three different questions. Input privacy: Can the computing provider process the request without learning the confidential input?

Program protection: Can somebody use the software without obtaining its internal design?

Verification: Can the customer, or another checker, test whether the claimed calculation was performed correctly?

Satisfying one does not automatically satisfy the others. A system might conceal a file but return an unreliable answer. It might return a valid result while exposing the information used to obtain it. And hiding a program is not the same as keeping its inputs secret.

This is why “secure quantum computing” is too broad a description to be useful on its own. A buyer needs to ask what is hidden, from whom, and what can actually be checked.

Checking without opening the private input

The paper’s technical bridge is quantum fully homomorphic encryption: a framework for evaluating quantum computations on encrypted classical information. Its construction adds a publicly checkable proof of evaluation. The encrypted messages and checking procedure are classical; the evaluation still requires quantum computation.^[2]

Public verification does not mean publishing the private input. It means checking need not depend on a secret verification key held by one designated checker.

The authors also use the construction to protect descriptions of certain quantum programs. The protected description is ordinary digital information, rather than a quantum state that must be stored or transmitted as such.^[2]

For a newcomer, the useful distinction is between an answer and evidence about that answer. A promise that the provider is trustworthy asks the customer to rely on the provider. A verification protocol specifies what evidence is accepted and under which assumptions.

It is not a guarantee that the original research question was sensible. Correctly executing a flawed model would still leave the user with a flawed model.

The small print is part of the result

The construction concerns pseudo-deterministic quantum circuits: roughly, computations that return the same classical answer for a given input except with negligible error. It does not establish protection for every possible quantum program.

Its security argument also uses a classical oracle, an idealised component, and assumes a mathematical problem called Learning With Errors remains hard for quantum computers. The paper leaves extending public verification to arbitrary output distributions among its open questions.^[2]

These restrictions prevent an easy leap from theorem to purchasing decision. A mathematical construction does not supply a measured deployment cost, an operational service or a benchmark on today’s hardware.

NTT itself presents the work as foundational research rather than a product.^[1]

The result’s value is therefore a clearer statement of what might fit together: private inputs, protected software and independently checkable computation. The practical milestone worth watching is evidence that a concrete implementation can preserve the relevant guarantees at a useful cost.

Until then, the best question is not “Is this quantum cloud secure?” It is “Which protection has actually been established?”