zkvmBlast finds seven bugs in Ethereum zkVM implementations

A valid Ethereum block can run cleanly on every node and still hit a wall: the machine asked to prove it may crash on an obscure instruction. zkvmBlast, a new differential fuzzer for RISC-V zero-knowledge virtual machines, is designed to find that kind of failure before proving becomes part of Ethereum’s core path.

The tool takes one program, runs it across multiple zkVMs and a reference simulator, then reports disagreements. It looks for soundness bugs, where a prover could convince a verifier of something false, and completeness bugs, where a valid program cannot be proved. It also checks executor correctness: whether the zkVM actually follows the RISC-V specification, the instruction set it is supposed to implement.

The first campaigns found seven distinct completeness and correctness bugs across SP1, Pico, and OpenVM. Each was responsibly disclosed. The work also reproduced a known soundness bug in RISC0. The findings were made as Ethereum’s relationship with zkVMs is changing: systems once used mainly by rollups, bridges, or coprocessors are being considered for proving ordinary Ethereum blocks.

That shift raises the cost of a seemingly harmless failure. A completeness bug does not let an attacker forge a balance; it may simply prevent a legitimate execution from producing a proof. But if Ethereum requires blocks to carry validity proofs, an accidental crash—or a transaction deliberately crafted to trigger one—could stop proving and threaten the chain’s liveness. The Ethereum Foundation’s Ethproofs project tracks proof latency and cost for real mainnet blocks, while its zkEVM team also tests worst-case, adversarial blocks.

So what changes in practice? zkvmBlast gives developers a way to compare implementations against one another and against a RISC-V reference before those implementations carry protocol-level responsibility. It does not show that Ethereum is already proving every block, nor that the bugs found would halt the network today. It shows where the next engineering pressure point lies: making rare instructions and corner cases as dependable as the headline proving benchmarks.