# Aztec Discloses Critical Proving-System Flaw That Could Validate Invalid Transactions

Core contributors found the defect through internal AI-assisted auditing and told users to treat funds, applications and contract state on the Alpha V5 network as exposed until operators complete the required network actions.

- Published: 2026-09-16T05:44:37.392Z
- Canonical: https://polylog.news/crypto/2026-09-16/aztec-discloses-critical-proving-system-flaw-that-could-vali
- 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 (protocol blog)](https://aztec.network/blog/how-gas-works-on-aztec)

Aztec, the privacy-focused Ethereum layer-2 network, disclosed a critical vulnerability in the proving system behind its [Alpha V5 release](https://aztec.network/alpha-v5). According to the project's [disclosure](https://aztec.network/blog/alpha-v5-proving-system-vulnerability), an attacker could construct a proof that passes verification for a transaction the network should reject, producing a state transition outside the rules the protocol intends to enforce. Core contributors said they found the defect through internal auditing assisted by artificial intelligence (AI) tools. They advised users to treat funds, applications and contract state on V5 as exposed until incident response is complete and network operators carry out the required actions.

The failure class matters more than this particular network. In an optimistic rollup, validators re-execute transactions and can catch a bad state transition. In a zero-knowledge system, the proof is the only thing standing between a submitted transaction and accepted state, so a soundness defect in the proving stack is not caught by anyone re-running the computation. It is caught by auditing, formal verification, or by an attacker.

This is the second such episode for Aztec within months. When the project moved from V4 to V5, it told users to [withdraw before June 25](https://thedefiant.io/news/defi/aztec-v4-withdraw-june-25-v5-upgrade-security-vulnerability) because the upgrade had exposed critical flaws in the earlier release. Aztec has raised its bug bounty to [$2 million](https://aztec.network/blog/aztec-network-raises-total-bug-bounty-to-2-million) and says more than 234 researchers have engaged with the program on Cantina. The network remains in alpha, which limits the value at risk, and the contributors said further bugs should be expected as usage grows.

The disclosure can be read two ways. Disclosing a soundness bug found in-house, before an attacker used it, is the behavior a serious engineering organization should show. At the same time, each disclosure confirms that privacy rollups are shipping systems whose security nobody, including their authors, can yet fully verify, and that users who bridge assets into them are accepting that residual risk in exchange for confidentiality.

## What this means

Soundness bugs in zero-knowledge proving systems bypass the fallback that most rollup users implicitly rely on, because no validator re-executes the transaction to notice the state is wrong. That exposes anyone holding assets on a proof-only chain to a total-loss event rather than a partial one, and it pushes serious deployments toward proof-system diversity, escape hatches and delayed upgrade windows. AI-assisted auditing now serves both defenders and attackers. It found this defect before an attacker did, and it is the same capability attackers can use against published circuits.

## What to watch

- Whether Aztec publishes a full post-mortem naming the specific circuit or verifier component at fault, which determines whether other teams using the same proving stack need to audit their own code.
- Whether other zero-knowledge layer-2 networks add a second independent proof system or a user escape hatch, which would show the industry now treats proving bugs as an expected failure mode rather than an outlier.
- Total value bridged into privacy rollups after the disclosure, as a direct measure of whether users price soundness risk or ignore it.