Morning Edition · Saturday, September 26, 2026Published at 1:50 AM EDT · New York
An attacker could construct a proof that verifies for a transaction the network should reject, and contributors found the defect through internal artificial-intelligence-assisted auditing.

Aztec, the privacy-focused Ethereum layer-2 network, has disclosed a critical vulnerability in the proving system behind its Alpha V5 release. Core contributors identified the defect on July 27, 2026 through internal auditing assisted by artificial intelligence tools. The flaw allows an attacker to construct a proof that passes verification for a transaction the network should have rejected, which is a soundness failure rather than a bug in application code.
That distinction determines how serious the flaw is. On a zero-knowledge rollup, the verifier contract accepts state transitions because a proof says they are valid. Validators do not re-execute private transactions to check the result, so this kind of defect cannot be caught by the re-execution check that optimistic rollups depend on. Aztec said funds, applications and contract state on Alpha V5 should be treated as exposed to a protocol-level failure until contributors finish incident response and network operators complete the required actions. Internal and external human audits have concluded, artificial-intelligence-assisted review continues, and reviewers report no other critical or high-severity findings in V5 at this time. The repair is scheduled for V6.
Aztec has been shipping quickly into this Alpha phase, including a live on-chain version of the hidden-information game Dark Forest built to demonstrate private state. Alpha networks exist precisely so that this class of defect surfaces before real value depends on it. The finding came from the team's own tooling rather than from an external audit. That shows how difficult it is to review a proving system from outside the team that built it.
Aztec, which converts a severe defect into evidence that its embargo-and-disclose process works before real value depends on the network, and the vendors selling artificial-intelligence-assisted auditing and formal verification to every other zero-knowledge team.
Part of a tracked trend
Proving-System Bugs Become a Distinct Rollup Risk
Soundness defects in zero-knowledge proving systems will keep surfacing as a risk class separate from smart-contract exploits, because validator re-execution — the fallback most rollups rely on — cannot catch them, forcing teams into embargoed disclosure timed to upgrades and pushing users toward proof-system diversity and escape hatches.
Start a discussion in Townsquare.
More from this edition
Every material detail, including the July 27 discovery date, the claim that the tooling found it first, and the statement that no other critical findings exist, comes only from Aztec's own disclosure, and contributors concede they cannot establish whether anyone exploited the flaw during the two months before publication.
An open-source-intelligence read of how likely this story is true with its real nuance, not a judgment of any outlet. It assesses the claim, weighing independent and adversarial reporting. How we label confidence.
What this means
The exposed party is anyone holding balances or deployed contract state on an Alpha network whose safety rests on a single proving stack, and the mechanism is that proof verification substitutes for re-execution, so a defect in the prover becomes a defect in the ledger itself. The precedent is Zcash, which discovered a subversion-of-soundness flaw in its original proving system in 2018, patched it quietly ahead of disclosure, and found no evidence of exploitation. Aztec's handling follows the same pattern of embargo, fix, then publication. The likelier outcome over the coming months is that privacy-focused rollups compete on prover assurance, meaning multiple independent proving implementations, formal verification of circuit constraints, and working escape hatches, rather than on transaction throughput or fee level. What would disprove that read is a repeat soundness disclosure that causes a real loss on a live network, which would push builders toward simpler designs with smaller trusted circuits instead.
What to watch
Observations to monitor, not financial advice.
Synthesized from: Aztec Network · Aztec Network · Aztec Network
Comments
0No comments yet.