Aztec Foundation co-founder and CEO Zac Williamson said he “can’t rule out the possibility that bugs will be found by more advanced AI models” in Aztec Tierra (Alpha V6), the latest version of the Ethereum privacy network, which launched on Nov. 3.

Williamson said on Unchained Premium, in an interview recorded Nov. 9, that the AI audits Aztec ran on the previous version turned up “some criticals” in its cryptographic libraries. Aztec had already disclosed in August that internal AI-assisted auditing found one critical flaw in the V5 proving system, a bug that could let a forged proof pass, and said it could not tell whether anyone had used it. An Oct. 15 post he co-wrote said Aztec’s AI audit harness had found critical vulnerabilities, including some that human audits missed, and that V6 patches them. Tierra fixes the V5 flaw, according to Aztec’s launch post, which also announced an AI testing harness and formal verification (mathematically proving code does what it should) of core cryptographic pieces.

On the show, Williamson described those audits as running 24/7 on Aztec’s own harness. He framed the stakes as a world where “the marginal cost of cyber attacks is falling to zero” and attackers can deploy “millions of AI bots.”


This story is brought to you by Unchained Premium.

Subscribe to get exclusive interviews, a subscriber-only chat group, and show transcripts.

Upgrade


Only Part of the Code Is Formally Verified

He also said “we’ve not fully formally verified our entire protocol yet,” and that only “the most security critical parts” have been. After the AI found the critical bugs, the team formally verified those subsystems, which he said implies “the bug that AI found was the last bug in that particular subsystem.” Williamson said the team rarely hits model refusals because it hunts cryptography and ordinary software bugs, and it runs uncensored open-source models “just to double-check.”

Upgrades as Hard Forks

Asked by host Laura Shin about the Zcash bug that Unchained reported in June, Williamson said that in privacy networks “bugs can often lead to double spend” attacks (spending the same funds twice) “where you can’t actually observe that this has happened.” Zcash’s own advisory said its turnstile accounting capped how much value could leave the affected pool.

Williamson said “Zcash are absolute best in class when it comes to managing this risk” and that Aztec copies the turnstile idea: “when we do major network upgrades, we do them effectively as complete hard forks,” so users “need to withdraw from the previous old version of Aztec” and deposit again. A hidden double-spend would show up in those withdrawals for tokens that live on Ethereum, he said, and “it can’t extend to the new version of Aztec.”

ETHBEAT (formerly L2BEAT), a site that tracks Ethereum layer-2 networks, lists V6 as a fresh start with no state carried over from V5, and rates it Stage 2, its top tier. Williamson added that “we don’t have a security council like other networks.” Aztec Labs relaunched its zk.money payments wallet on the network in September.

Williamson said “the long-term solution is formal verification.”

Watch the full conversation on Unchained Premium: Premium members get the complete interview with Aztec Foundation’s Zac Williamson on whether private protocols can keep North Korea out. Watch it here.