# Aztec Discloses Critical Soundness Flaw in Alpha V5 Proving System and Tells Teams to Halt Deployments

The privacy layer-2 says an attacker could have constructed a valid-looking proof for a transaction the network should reject, and that chain history cannot show whether anyone did.

- Published: 2026-10-02T06:09:12.150Z
- Canonical: https://polylog.news/crypto/2026-10-02/aztec-discloses-critical-soundness-flaw-in-alpha-v5-proving
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [Aztec Network (disclosure)](https://aztec.network/blog/alpha-v5-proving-system-vulnerability), [Aztec Network (Alpha V5)](https://aztec.network/alpha-v5), [The Defiant](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability), [Aztec Network (bug bounty)](https://aztec.network/blog/aztec-network-raises-total-bug-bounty-to-2-million)

Aztec, the privacy-focused Ethereum layer-2 network, has disclosed a critical vulnerability in the proving system behind its [Alpha V5 release](https://aztec.network/blog/alpha-v5-proving-system-vulnerability). According to the disclosure, an attacker could construct a proof that passes verification for a transaction the network is supposed to reject, and an accepted transaction of that kind could move the chain into a state its own rules forbid.

Core contributors say they found the defect on 27 July 2026 through internal auditing assisted by artificial intelligence, and that the fix is scheduled for the next release, V6. The disclosure is unusually candid about what cannot be established: the affected system does not record enough information to separate ordinary accepted transactions from transactions accepted through the flawed proving path, so historical activity cannot confirm whether anyone exploited the bug or size the damage. Teams preparing V5 deployments are told to pause, and teams already live are told to review how they would limit user exposure, isolate the affected functionality, and migrate users to fresh contracts.

This is the second critical finding in Aztec's alpha cycle. In June the project told [users of the previous version to withdraw funds](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability) ahead of the V5 upgrade after a flaw surfaced in Alpha V4. Aztec has also [raised its bug bounty to $2 million](https://aztec.network/blog/aztec-network-raises-total-bug-bounty-to-2-million), a sign of how much it is paying to find defects that conventional review misses.

The failure class matters beyond one network. A soundness bug in a proving system is not a contract bug that validators can catch by re-executing transactions. If the verifier accepts a false statement, every downstream assumption built on that proof inherits the error. L2Beat puts total value secured across Ethereum layer-2s at $51.10 billion, much of it dependent on proof systems whose correctness users cannot independently check.

## What this means

The exposed parties are builders and depositors on proof-verified privacy networks, and the channel is verifier trust rather than contract logic: when a proving system can be convinced of a false statement, there is no re-execution fallback and no reliable forensic record, which is exactly what Aztec concedes here. Aztec wants to keep alpha-stage credibility by disclosing early, and it can, because alpha networks carry little value. The precedent is its own June Alpha V4 incident, which ended in a mandatory user withdrawal window before an upgrade. The likelier outcome over the coming months is more embargoed proving-system disclosures across zero-knowledge rollups timed to scheduled upgrades, with escape hatches, multi-prover setups and staged value caps becoming standard pitches to institutional users rather than optional extras. That read would weaken if V6 ships with an independent third-party soundness audit and no comparable defect appears at other zero-knowledge teams through the next upgrade cycle.

## What to watch

- Whether Aztec publishes the technical specifics of the flawed proving path alongside the V6 fix, which would let other teams check their own circuits for the same error class.
- Whether other zero-knowledge rollups add a second, independent prover or a user-triggered withdrawal path, the practical test of whether proof diversity is being adopted or only discussed.
- Whether institutional pilots on privacy layer-2s slow down after this disclosure, which would show that confidential execution still carries an unpriced correctness risk.