AivexaNewsSearch
AI news for builders and product teamsChecked every hour

Proof-based general-purpose computing with Magnetar

Collected Oct 15, 2026

Microsoft Research has introduced Magnetar, a zero-knowledge virtual machine (zkVM) that proves the execution of programs written in Rust, Go or C and, alongside the output, emits a short cryptographic proof that the program ran correctly. The work is presented as the latest output of more than a decade of the lab's research into making zero-knowledge proofs practical, and the team reports it makes major progress on three fronts at once: proving speed, usability for ordinary programs, and long-lived cryptographic security.

A zkVM proves the execution of an ordinary processor rather than a hand-built arithmetic circuit. Magnetar proves RISC-V, an open instruction set architecture with mature compilers for mainstream languages, so a developer compiles as usual and the execution is proven as written. That choice matters because the traditional alternative, translating a program by hand into a circuit, is slow and fragile in practice: every application change requires a new circuit, and the translation itself is a frequent source of security bugs. Magnetar's framing is that a business policy, a data pipeline, or an EVM interpreter is just a program, so the engineers who maintain it can also produce proofs about it.

The headline performance figures are specific. Magnetar proves between 7.2 and 8.7 million processor instructions per second on a laptop. Across 15 benchmark programs, ranging from cryptographic hashing to token transfers inside a small EVM interpreter, it is reported to be 10 to 39 times faster on average than the fastest existing zkVMs on the same machine, and more than 300 times faster than two others on the token-transfer benchmark. An optimized configuration reaches 9.5 to 12.8 million cycles per second, which the team says raises each reported speedup by roughly 1.3 times. Proofs verify in about 12 milliseconds.

The security claim is the part with the longest horizon. Magnetar depends on a standard hash function such as SHA-256 or SHA-3 and requires no elliptic curves, no lattices, no special-purpose hash functions, no trusted setup and no unproven conjecture about error-correcting codes. That removes several categories of risk that hang over other systems: elliptic-curve proofs fall to a sufficiently powerful quantum computer, trusted setups carry secret material whose leak would allow forged proofs, and newer fast hash functions have not had decades of public scrutiny. For context, hash-based constructions are the basis of the stateless signature schemes NIST standardized as its conservative post-quantum option, so the assumption Magnetar leans on is among the best studied in cryptography. Microsoft's broader Quantum Safe Program aims to move its products and services to post-quantum cryptography, a goal the post says was recently accelerated to 2029.

One design decision has a second payoff worth noting for developers. Verifying a Magnetar proof consists mostly of evaluating standard hash functions and binary-field arithmetic, which are exactly the operations Magnetar proves efficiently. Verification is therefore an ordinary workload for the system itself. That is what enables proving computations of any size in pieces and then combining them, and it is what would let Magnetar prove a long-running system such as a ledger or a blockchain. In practice this is the composition property that makes recursive or incremental proving tractable; the report frames it as an outgrowth of the same cryptographic choice rather than an added feature.

The system sits at the end of a long research line the post lays out. Spartan, published in 2019, returned to the sum-check protocol and made it work for general-purpose programs without a trusted setup, and its flexibility later allowed instantiation with hash functions alone. Phalanx (2021) exploited the repetition in a processor's instruction cycle to speed proving by an order of magnitude. The Nova, HyperNova, NeutronNova, Neo and SuperNeo line showed that stopping proving early and folding partial results together makes proving computations of any size far cheaper. Brakedown (2021) introduced a fast hash-based commitment scheme combined with Spartan, producing the first Spartan variant resting on standard hashes. Spice, Lasso, Nebula, Twist and Shout attacked the cost of proving memory operations, and Jolt showed how to combine those with Spartan to prove a complete processor. Vega applied the line of work to digital identity, proving facts from a government credential in under a tenth of a second. BinarySpartan brought Spartan to binary fields, where processor logic and standard hashes are cheap to express, leaning on Ligerito for shorter proofs and on ring switching for efficient commitment to bit vectors.

A likely trade-off, as I read the claims, is that the comparison base is other zkVMs on the same machine rather than the cost of running the program normally. Producing a proof still imposes substantial overhead relative to plain execution, and the reported instruction rates describe a laptop-class prover. On the security side, the absence of a trusted setup and exotic assumptions is a real simplification for auditors and for migration planning, but hash-only systems have historically paid for that conservatism in proof size and prover cost; the post emphasizes speed and short verification times but does not present proof sizes against competitors in the excerpted material. Whether the hash-based approach stays competitive at very large workloads is the empirical question the benchmarks begin to answer.

Why it matters: Teams that want verifiable computation currently face a choice between hand-circuitizing their logic and accepting weaker cryptographic assumptions, and Magnetar's reported combination of RISC-V generality, double-digit speedups and hash-only security narrows that gap. If the numbers hold up under independent testing, product teams could attach proofs to policy-bound agent actions, financial processing or data pipelines using code they already write in Rust, Go or C, which is an inference about adoption rather than a stated result. The immediate change for developers is conceptual: proving stops being a cryptography specialty and becomes a build step, with the remaining cost shifting to how much overhead a given use case will tolerate.

Read at Microsoft Research

Based on reporting from the original publisher. Visit the source for full context and later updates.

Publisher excerpt

Zero-knowledge proofs can verify computing results without rerunning workloads or exposing sensitive data. Magnetar helps bring them into the mainstream with support for standard programming languages, major speed gains, and quantum-resistant security. The post Proof-based general-purpose computing with Magnetar appeared first on Microsoft Research .