# Aztec Discloses Critical Proving-System Flaw in Its Live Privacy Network

Contributors say an attacker could craft a proof that the network would wrongly accept, a soundness failure that validator re-execution cannot catch. The disclosure comes as Ethereum's own zero-knowledge benchmark shows a 52-bit shortfall against its December target.

- Published: 2026-08-22T05:43:59.635Z
- Canonical: https://polylog.news/crypto/2026-08-22/aztec-discloses-critical-proving-system-flaw-in-its-live-pri
- 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), [CryptoSlate](https://cryptoslate.com/ethereum-researchers-are-racing-to-close-a-zkevm-security-gap-before-december/)

Core contributors to Aztec, the privacy-focused layer-2 network built on Ethereum, have [disclosed a critical vulnerability in the proving system](https://aztec.network/blog/alpha-v5-proving-system-vulnerability) that secures its Alpha V5 release. The defect is one of soundness: an attacker may be able to construct a zero-knowledge proof that the verifier accepts for a transaction the network should reject. Contributors said funds, applications and contract state on V5 should be treated as exposed to a protocol-level failure until incident response finishes, and that the flaw was found through internal auditing assisted by artificial-intelligence (AI) tools rather than by an outside attacker.

This is the second proving-system defect Aztec has disclosed this year. A comparable critical flaw in Alpha V4 forced contributors to tell users to withdraw ahead of the [V5 upgrade](https://aztec.network/alpha-v5), which cut private-transaction proving costs and was activated through on-chain governance. Contributors say human audits, internal and external, have completed and that reviewers have not identified other high-severity or critical issues in V5, with AI-assisted review continuing.

Soundness bugs form a separate category of risk. A smart-contract exploit drains a specific application, and honest nodes can still see what happened by replaying the transactions. A broken proof system corrupts the mechanism by which the chain decides what is true, and re-execution by validators, the fallback most rollups rely on, cannot detect a forged proof it never sees. That is why disclosure of these defects tends to be embargoed and timed to an upgrade.

Ethereum faces a version of the same problem at its base layer, the underlying blockchain that layer-2 networks like Aztec settle onto. The Ethereum Foundation set a target of 128-bit provable security for a production zero-knowledge Ethereum Virtual Machine (zkEVM) by the end of 2026, with proofs under 300 kilobytes and formal arguments that the recursion architecture is sound. CryptoSlate [reports that the better.codes contest](https://cryptoslate.com/ethereum-researchers-are-racing-to-close-a-zkevm-security-gap-before-december/), which scores live security certificates rather than abstract claims, currently shows a 63.99-bit lower certificate against a 116.13-bit upper certificate for the koalaIRS12 profile. The 52.14-bit interval is unresolved evidence, not proven weakness, but it is the distance researchers must close before December, and the score covers only one profile.

The two disclosures together show where blockchain risk has shifted. Execution and data availability are largely solved engineering problems. Whether a proof actually proves what it claims is not, and the industry is now measuring that question in bits rather than asserting it in prose.

## What this means

Rollups and privacy chains ask users to trust a mathematical argument instead of a set of operators, and a soundness defect makes that argument false without any visible on-chain symptom. Holders of assets on proof-secured networks, and the applications built on them, carry a risk that contract audits and bug bounties are not scoped to find. The direct consequence concerns where capital sits: value tends to stay on chains with forced-exit mechanisms, proof-system diversity or validator fallbacks, and to leave networks where a single prover is the only safeguard.

## What to watch

- Whether Aztec publishes a full technical post-mortem naming the affected circuit component, which would tell builders on other zero-knowledge stacks whether the defect is specific to Aztec or shared across systems using similar recursion.
- Whether better.codes certificate scores for the koalaIRS12 profile narrow in the autumn, because a gap that stays wide into November would force Ethereum's zkEVM teams to either slip the December milestone or accept a lower security bound.
- Whether other zero-knowledge layer-2 teams begin disclosing internally found proving bugs on their own schedule, which would show AI-assisted auditing is now surfacing defects faster than attackers find them.