# Aztec Discloses a Proving-System Flaw That Let an Invalid Transaction Pass Verification

The privacy network says its own contributors found the defect through internal AI-assisted review, and that validator re-execution, the fallback most rollups rely on, does not catch this class of bug.

- Published: 2026-09-14T05:49:22.021Z
- Canonical: https://polylog.news/crypto/2026-09-14/aztec-discloses-a-proving-system-flaw-that-let-an-invalid-tr
- Publisher: Polylog (Crypto desk)
- Section: crypto
- Sources: [Aztec Network](https://aztec.network/blog/alpha-v5-proving-system-vulnerability), [Aztec Network](https://aztec.network/alpha-v5), [Aztec Network](https://aztec.network/blog/how-gas-works-on-aztec), [The Defiant](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability)

Aztec, the zero-knowledge privacy layer-2 network on Ethereum, has [disclosed a critical vulnerability in the proving system of its Alpha v5 release](https://aztec.network/blog/alpha-v5-proving-system-vulnerability). By the team's description, an attacker could construct a proof that passes verification for a transaction the network should reject. Core contributors identified the defect on 27 July 2026 through internal auditing assisted by artificial intelligence (AI) tooling, and the fix is scheduled for the next version rather than an emergency patch.

The distinction matters. A smart-contract bug lets an attacker misuse rules that the chain still enforces correctly. A soundness defect in the proving system breaks the enforcement itself, which means the chain can be made to accept state it would otherwise reject. Aztec has stated that this class of flaw is not caught by public re-execution by the validator committee, the check that ordinary optimistic and hybrid designs depend on.

This is the second such disclosure in six months from the same team. Aztec [reported a critical flaw affecting the v4 proving system as a whole in March](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability), warned that exploitation could allow theft of user funds, told v4 users to withdraw, and withheld technical details until the v5 upgrade shipped. Embargoed disclosure timed to an upgrade is now the standard practice for proving-system bugs across the zero-knowledge sector, because publishing the defect before the fix exists would give an attacker a working method.

Two readings compete, and both are available from the same facts. Aztec presents the finding as the intended output of an alpha network, meaning a live system carrying deliberately limited value while the proving stack is hardened. Critics of the rollup model see a pattern in which the security case for a chain rests on cryptography that only a small number of specialists can review, leaving users to trust a disclosure process instead of an independent check.

## What this means

Proving soundness is the assumption underlying every zero-knowledge rollup and every privacy chain that hides transaction contents, because a forged proof looks identical to a valid one to anyone downstream. Users of proof-only systems cannot detect this kind of failure through block explorers or re-execution, so they depend on the quality of one team's review process and the timing of its disclosures. Chains that keep an escape hatch or a second, independent proving implementation reduce that exposure. The sector-wide consequence is slower institutional adoption of proof-only privacy platforms until independent verification of proving systems matures.

## What to watch

- Whether Aztec publishes the full technical detail of the v5 defect once the fix ships, which is the test of whether its stated transparency policy survives a second incident.
- Whether other zero-knowledge rollups adopt multiple independent proving implementations, since diversity is the only structural defense against a single soundness bug.
- Whether AI-assisted internal audits keep producing this class of finding, which would mean disclosure timing, not new attacker research, is setting the patch schedule.