Aztec Discloses a Critical Flaw in the Proving System Securing Its Privacy Network

Aztec, the privacy-focused Ethereum layer-2 network, disclosed a critical vulnerability in the proving system of its Alpha V5 release. Core contributors identified the defect on 27 July 2026 during internal auditing assisted by artificial intelligence tools. The network says an attacker could build a proof that passes verification for a transaction the protocol should reject, producing a state transition outside the rules V5 is meant to enforce.

That is a soundness failure, not a bug in an application contract. In a zero-knowledge rollup, the proof is the rule. Optimistic systems and ordinary chains rely on other nodes re-executing transactions to catch invalid state, and a privacy network cannot fall back on that, because the transaction contents nodes would need to check are the very thing being hidden. Aztec told users to treat funds, applications and contract state on V5 as exposed to a protocol-level failure until contributors finish incident response and operators carry out the required network actions. The permanent fix is planned for V6, due later in 2026.

This is the second consecutive version affected. Aztec disclosed a critical vulnerability in Alpha V4 in March 2026, saying it touched the proving system as a whole and could lead to theft of user funds, and The Defiant reported that users were told to withdraw before 25 June ahead of the disclosure becoming public. The pattern is consistent with a deliberate policy: hold the details until an upgrade path exists, then publish.

The disclosure lands while V5 carries live applications, including a fully on-chain version of the strategy game Dark Forest that uses private state as a game mechanic. Aztec says reviewers have found no other high-severity or critical V5 defects, that internal and external human audits are complete, and that machine-assisted review continues.

The method of discovery matters as much as the bug. The defect was found by an audit process using AI tooling, on code that had already passed human review. Proving systems are mathematically dense, thinly staffed globally, and now reviewed by automated tools that can run without stopping.