News analysis ·

Aztec Tierra: why a privacy layer 2 upgrades by hard fork

Aztec Tierra (Alpha V6) went live with AI audits and partial formal verification. We explain its hard fork upgrade model and who it affects.

Black-and-white photo of a row of subway turnstiles with card readers and an iron gate, set in a dark frame on a cream page under a burgundy rule.
Photo: Tdorante10 / Wikimedia Commons, CC BY-SA 4.0 (cropped, black and white)

Aztec Tierra, also called Alpha V6, went live on Nov. 3 as the newest release of the Aztec privacy network on Ethereum. In an interview with Unchained, Zac Williamson, who co-founded the Aztec Foundation and runs it as CEO, said he cannot rule out that stronger AI models will still find bugs in it. This article looks at how Aztec tries to contain bugs it cannot see, and why each major upgrade forces users to move their funds.

The change in one paragraph

Tierra replaces the previous version, V5, rather than updating it in place. According to Aztec's launch post, as reported by Unchained, the new release fixes a critical flaw found in the V5 proving system, adds an AI testing harness and formally verifies core cryptographic components. Formal verification means writing a mathematical proof that a piece of code does exactly what its specification says. The catch, in Williamson's own words on Unchained Premium, is that only the most security critical parts are verified so far, not the whole protocol.

What is a layer 2, and why privacy makes bugs harder to spot

A layer 2 is a separate network that runs on top of a base chain such as Ethereum. Users deposit tokens into a contract on Ethereum, transact on the layer 2, and later withdraw. The layer 2 keeps its own records and proves to Ethereum that they are valid.

On a public network, a broken balance tends to show up because anyone can add up the numbers. A privacy network hides amounts and owners on purpose. Williamson warned that here a bug can allow a double spend, spending the same funds twice, without anyone seeing it directly. He pointed to a Zcash bug that Unchained reported in June.

Under the hood: the turnstile and the hard fork

Zcash's answer to hidden inflation is a turnstile. According to Zcash's advisory, its turnstile accounting put a ceiling on the value that could exit the affected pool. Williamson called Zcash best in class at managing this risk, and said Aztec borrows the idea.

In Aztec's version, every major upgrade is run as a complete hard fork. A hard fork is a break in the rules: the new network does not accept the old one's history. Users have to withdraw from the old version and deposit into the new one. Here is how the moving parts fit together:

  • Deposits of Ethereum-native tokens sit in a contract on Ethereum, where the totals are public.
  • When an upgrade ships, users exit the old version, and every exit is checked against that public pool.
  • If someone had quietly created extra value through a bug, the exits would try to take out more than was deposited, and the gap would surface there.
  • The new version starts clean. ETHBEAT, formerly L2BEAT, treats V6 as a clean restart that carries no state over from V5.

Williamson said a concealed double spend would surface in those withdrawals and could not carry into the new version. Each upgrade doubles as an audit of the old network's books. The trade-off is friction: users must act at every major release.

AI audits and formal verification

The second line of defence is finding bugs before attackers do. In August, Aztec disclosed that its in-house AI-assisted auditing had caught a critical flaw in the V5 proving system that could let a forged proof be accepted. A proof is the cryptographic receipt a layer 2 sends to Ethereum; a forged one could make false records look valid. Aztec said it could not tell whether the flaw had been used.

An Oct. 15 post co-written by Williamson said the AI harness found critical vulnerabilities, some missed by human auditors, and that V6 fixes them. Williamson told Unchained the audits run 24/7 on Aztec's own harness, and that the team also runs uncensored open-source models as a cross-check. Once the AI flagged those bugs, engineers formally verified the affected subsystems, which he said implies no further bugs remain in them.

That claim is narrow by design. A formal proof covers only the code it is written for; the rest still depends on testing and audits, which can miss things. Williamson summed up the direction plainly: "the long-term solution is formal verification."

Who is affected

Users carry the most direct burden. Because each major upgrade is a full hard fork, they have to withdraw from the old version and redeposit into the new one. Apps building on Aztec, such as the zk.money payments wallet that Aztec Labs relaunched on the network in September, sit on the same network and are exposed to the same upgrade model.

There is also a governance point. Williamson said Aztec has no security council, unlike other networks. A security council is a small group that can push emergency fixes. He said formal verification is the long-term answer instead. ETHBEAT gives V6 Stage 2, the top tier on its scale.

Status and what to watch

Tierra is live and, by the team's own account, only partly formally verified. The remaining risk is the one Williamson named: newer AI models may find flaws current ones missed. Watch how much more of the protocol gains formal proofs, and how smoothly the next migration runs.

Sources

AI wrote this summary from the primary documents and reports cited above. It went live automatically after checks that figures match those sources and nothing is copied; there was no human review. It explains a protocol or rule change and is not investment advice.

← All posts