# Aztec Discloses Critical Soundness Flaw in the Proving System Running Its Live Privacy Network

Core contributors discovered on 27 July that an attacker could build a proof that passes verification for a transaction the network should reject. Internal review assisted by artificial intelligence found the bug, not an outside attacker.

- Published: 2026-09-03T05:55:38.226Z
- Canonical: https://polylog.news/crypto/2026-09-03/aztec-discloses-critical-soundness-flaw-in-the-proving-syste
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [Aztec Network](https://aztec.network/blog/alpha-v5-proving-system-vulnerability), [Aztec Network (Alpha V5)](https://aztec.network/alpha-v5), [Aztec Network (Dark Forest)](https://aztec.network/blog/dark-forest-aztec-game-goes-live)

Aztec, the privacy-focused layer 2 network built on Ethereum, has [disclosed a critical vulnerability](https://aztec.network/blog/alpha-v5-proving-system-vulnerability) in the proving system that secures its live [Alpha V5](https://aztec.network/alpha-v5) network. Aztec says core contributors identified the defect on 27 July 2026 during internal auditing assisted by artificial intelligence, and that an attacker could construct a proof that passes verification for a transaction the network should have rejected.

That is a soundness failure, a different category of problem from a bug in an application contract. A zero-knowledge proof system is designed to make invalid state transitions impossible to prove. When the proof system itself accepts an invalid statement, the network's verifier has no independent way to catch the error, because the design's purpose is to avoid having other participants re-execute the hidden transaction. On a privacy network, the encrypted contents weaken that safeguard even further.

The disclosure follows the same pattern Aztec set earlier this year. In June the team [told Alpha V4 users to withdraw funds before 25 June](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability), because putting the V5 upgrade to an on-chain governance vote would force public disclosure of the critical proving bugs still present in V4. Aztec has [raised its bug bounty to 2 million dollars](https://aztec.network/blog/aztec-network-raises-total-bug-bounty-to-2-million) and says reviewers have not identified other high-severity or critical defects in V5.

The timing matters because applications are moving onto the network. Aztec has been publicising live deployments, including [a version of the on-chain game Dark Forest](https://aztec.network/blog/dark-forest-aztec-game-goes-live). Aztec has not reported any theft of user funds tied to the V5 flaw.

## What this means

Every rollup that replaces re-execution with a proof depends entirely on the correctness of its prover, a dependency users cannot audit themselves. Holders of assets on any zero-knowledge network are exposed through a single channel: if the verifier accepts a false proof, balances can be created or moved without any contract bug ever appearing. That leaves users with only two practical checks on safety, namely whether an unforced exit route exists and how quickly the team discloses problems it finds. The offsetting factor is that machine-assisted review is now catching these defects before attackers do, which is why disclosures like this one are coming from operators rather than from reports of stolen funds.

## What to watch

- Whether Aztec's next governance vote includes a fix and forces another public disclosure window, which would confirm that transparency rules, not attacker pressure, now set the patch calendar for privacy networks.
- Whether other zero-knowledge networks begin publishing similar internal audit findings, which would show the problem is industry-wide rather than specific to one prover implementation.
- How much value users leave on privacy layer 2s while a known prover defect is outstanding, a direct measure of whether depositors price soundness risk at all.