# Aztec Discloses a Proving-System Flaw That Could Let Invalid Transactions Pass Verification

Core contributors found the defect through internal artificial-intelligence-assisted auditing on 27 July 2026, cannot rule out prior exploitation, and have deferred the fix to the next network version.

- Published: 2026-09-25T05:49:08.849Z
- Canonical: https://polylog.news/crypto/2026-09-25/aztec-discloses-a-proving-system-flaw-that-could-let-invalid
- 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/dark-forest-aztec-game-goes-live)

Aztec, the privacy-focused layer-2 network on Ethereum, has [disclosed a critical vulnerability in the proving system](https://aztec.network/blog/alpha-v5-proving-system-vulnerability) behind its Alpha V5 release. According to the disclosure, an attacker may be able to construct a proof that passes verification for a transaction the network should reject, which would produce a state transition outside the rules the protocol intends to enforce. Contributors identified the defect on 27 July 2026 through internal auditing assisted by artificial intelligence, and they state they cannot determine whether anyone exploited it beforehand.

Aztec tells users to treat funds, applications and contract state on V5 as exposed to a protocol-level failure until incident response finishes and node operators complete the required network actions. The fix is planned for V6. Reviewers say internal and external human audits have completed without finding other critical or high-severity issues in V5, and that machine-assisted review continues.

This is the second consecutive Alpha version to carry a critical proving-system defect. Aztec disclosed a comparable flaw in Alpha V4 in March, with the patch shipped in the V5 release. The pattern matters because a soundness bug in a zero-knowledge proving system (cryptographic code that lets a network confirm a transaction is valid without re-executing it) is a different risk class from a smart-contract bug. Validators cannot catch it by re-executing transactions, since the fraudulent proof verifies correctly by construction. The network's [Alpha V5 program](https://aztec.network/alpha-v5) has meanwhile been onboarding private applications, including [a version of the on-chain game Dark Forest](https://aztec.network/blog/dark-forest-aztec-game-goes-live), so the exposed state is no longer purely experimental.

There is a second point worth noting. The defect was found by an internal audit process using artificial intelligence, the defensive counterpart to an argument now common in crypto security, where the same capability said to threaten cryptographic assumptions is also accelerating the discovery of implementation errors before attackers can exploit them.

## What this means

Rollups and privacy networks that rely on validity proofs have no fallback when the prover itself is unsound, because re-execution by honest validators cannot detect a proof that verifies. Users of Aztec's Alpha V5 carry direct exposure to any state transition an attacker could have forged, and the broader cost lands on every zero-knowledge network competing for institutional confidentiality work, because buyers now have a second concrete instance of the risk to price. The competing outcome is that repeated self-disclosure before exploitation builds credibility, which would favor teams with mature audit programs over faster shippers.

## What to watch

- Whether Aztec publishes evidence that the flaw was never exploited, which decides whether this becomes a disclosure story or a loss story.
- The V6 release timing and whether operators are required to halt or migrate state, which shows how much user disruption the fix demands.
- Whether other zero-knowledge rollups add proof-system diversity or escape hatches in response, the practical mitigation for soundness risk.