Microsoft Research Unveils Magnetar, a Post-Quantum zkVM

Microsoft Research has unveiled Magnetar, a system that produces a short mathematical proof that a computer program ran correctly. Anyone can check the proof quickly without running the program again. Microsoft says Magnetar is much faster than comparable systems and is designed to stay secure against future quantum computers, though the speed figures come from Microsoft’s own benchmarks and have not been independently tested.

The system is described in a Microsoft Research blog post by Srinath Setty, published on 15 October 2026, alongside a technical paper.

What Magnetar does

Magnetar is a zkVM, or zero-knowledge virtual machine. A virtual machine is software that behaves like a computer processor. A zkVM also generates a proof of what it computed, and can do so without revealing private inputs.

Magnetar proves the execution of programs written for RISC-V, an open processor instruction set. That means developers can write ordinary programs in Rust, Go or C, compile them and get proofs, rather than writing specialised cryptographic circuits by hand, Microsoft says.

The verifier’s job is cheap. According to the post, a Magnetar proof can be checked in about 12 milliseconds.

Microsoft’s performance claims

All of the following numbers are Microsoft’s own, measured by its researchers:

Across 15 benchmark programs, Magnetar was “10 to 39 times faster on average” than the prior fastest comparable systems, run on the same machine.
It proved 7.2 to 8.7 million processor instructions per second on a laptop.
On a token-transfer benchmark, it was more than 300 times faster than two other systems.
An optimised version reached 9.5 to 12.8 million cycles per second, roughly 1.3 times faster again.

The post does not name every competing system in its headline comparison, and no outside group has yet reproduced the results. Benchmarks published by a system’s own developers can depend heavily on how the tests are chosen and configured.

Built on hashes alone

Microsoft’s main security claim is about what Magnetar does not depend on. It uses only standard, widely studied hash functions, SHA-256 and SHA-3, with “no elliptic curves, no lattices, no special-purpose hash functions, no trusted setup.”

That matters for two reasons. Elliptic-curve cryptography, used by many existing proof systems, could be broken by a large enough quantum computer, while hash functions are thought to hold up far better. And a “trusted setup” is a one-off ceremony in some proof systems whose secret values, if leaked, would let someone forge proofs. Magnetar avoids that, and Microsoft says it also avoids relying on an unproven conjecture about error-correcting codes that some faster systems use.

Microsoft ties the work to its Quantum Safe Program, whose goal for moving to quantum-safe cryptography it says was recently accelerated to 2029.

Years of research behind it

Magnetar builds on a line of Microsoft Research proof systems that Setty and colleagues have published since Spartan in 2019, including Nova and others listed in the post.

Where it could be used

The post points to several uses. An AI agent acting on someone’s behalf “could attach a proof that each action complied with the policy it was given,” it says, and regulated financial institutions could prove that transactions were processed correctly. Because Magnetar proves large computations in pieces and then combines them, Microsoft says it can also prove a long-running system such as a ledger or a blockchain.

What’s still unknown

Microsoft has not announced a product, a release date, open-source code or licensing terms for Magnetar. Independent benchmarks against rival zkVMs, and peer review of the paper, are still to come.

What to watch

Watch for a public code release, third-party benchmarks and any sign of Magnetar being used in Microsoft products or by blockchain projects.

What this means

Proofs for ordinary code. Developers can prove Rust, Go or C programs ran correctly.
Big speed claims. Microsoft’s own tests show 10 to 39 times faster on average; others have not confirmed this.
Quantum-safe by design. It relies only on standard hash functions, with no trusted setup.