# Aztec Discloses a Soundness Flaw in Its Alpha V5 Proving System, With the Fix Deferred to V6

Internal AI-assisted auditing found that an attacker could build a proof that verifies for a transaction the privacy network should reject, the second critical proving bug Aztec has disclosed this year.

- Published: 2026-09-27T05:54:44.199Z
- Canonical: https://polylog.news/crypto/2026-09-27/aztec-discloses-a-soundness-flaw-in-its-alpha-v5-proving-sys
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [Aztec Network (vulnerability disclosure)](https://aztec.network/blog/alpha-v5-proving-system-vulnerability), [Aztec Network (Alpha V5)](https://aztec.network/alpha-v5), [Aztec Network (gas design)](https://aztec.network/blog/how-gas-works-on-aztec)

Aztec, the privacy-focused layer-2 network on Ethereum, has published a [disclosure covering a critical vulnerability in the proving system behind its Alpha V5 release](https://aztec.network/blog/alpha-v5-proving-system-vulnerability). Core contributors found the defect through internal auditing assisted by artificial intelligence tooling. In Aztec's description, an attacker could construct a proof that passes verification for a transaction the network is supposed to reject. The team says internal and external human audits had already been completed, that AI-assisted review continues, and that the correction is scheduled for the V6 release rather than a patch to the [live Alpha V5 network](https://aztec.network/alpha-v5).

This is the second such finding in Aztec's public record this year. In March, the team disclosed a critical proving-system vulnerability in Alpha V4 that could have allowed theft of user funds, and it asked V4 users to withdraw before the V5 governance vote made the details public. The pattern is the reason the network is labeled Alpha: contributors treat it as a test surface where the value at risk stays deliberately limited compared with [general-purpose rollups](https://aztec.network/blog/how-gas-works-on-aztec).

A soundness bug is a different risk class from a smart-contract exploit. Most optimistic rollups can fall back on re-execution by honest validators, which catches an invalid state transition. A validity proof system has no such backstop. If the verifier accepts a false proof, the chain treats an invalid transaction as final, and neither node operators nor users can spot the discrepancy from public data. That is why teams in this position disclose on their own schedule and time the details to an upgrade. It is also why Aztec's admission, that human audits missed what an automated review caught, matters to every zero-knowledge network shipping in production today.

## What this means

The exposure sits with anyone whose funds are secured by a single proving system with no independent re-execution path, which today means zero-knowledge rollups and the privacy networks built on them. The mechanism is simple: soundness failures are silent, so the loss surfaces after settlement rather than during it. Aztec benefits from finding this inside an alpha network with limited value at stake, and it also sets a disclosure norm that competitors will be measured against. The precedent worth naming is Aztec's own March V4 disclosure, which ended with users withdrawing ahead of a scheduled upgrade rather than with a theft. The likelier outcome over the next several months is more disclosures of this kind across zero-knowledge stacks, with teams moving toward multiple independent provers and user escape hatches as the standard answer, so long as the discoveries keep coming from defenders rather than attackers. If a soundness bug is instead exploited on a mainnet rollup holding real deposits, the market will start pricing proof-system risk into where it parks capital rather than treating it as an engineering detail.

## What to watch

- Whether Aztec ships V6 with the proving fix on its published timetable, and whether it asks Alpha V5 users to withdraw first, as it did before the V4 disclosure.
- How many other zero-knowledge rollups adopt multiple independent proving systems or a fallback verifier, the only structural defense against an accepted false proof.
- Whether AI-assisted auditing keeps producing findings that completed human audits missed, which would change how the market reads an audit report.