From Access Control to Encrypted Execution with Auryn Macmillan
Wednesday, 20 May 2026 · 4 min read · Listen to the episode ↗
Auryn Macmillan traces the evolution of Zodiac from the technical scaffolding behind the Gnosis DAO into a modular access control suite now securing roughly three billion dollars in assets across treasury managers, vault operators, and the Gnosis Pay debit card. He uses the Bybit exploit to argue that constraining routine operations to pre-defined roles rather than granting full Safe root access would have limited the blast radius.
Gnosis Guild spun out of Gnosis in 2023 and operates as an independent venture studio. Zodiac originated from the technical work of designing the Gnosis DAO and evolved into a modular access control suite built on top of Safe, serving treasury managers, vault managers, and consumer products including the Gnosis Pay debit card. It secures approximately 3 billion dollars in assets today, up from roughly 600 million dollars in 2022, with around 10 billion dollars transacted in total. The majority of transaction value comes from noncustodial treasury and vault activity, while Gnosis Pay cards account for most transactions by count. Zodiac has not experienced any major exploits, and most reports from its bug bounty program have been informational rather than significant vulnerabilities.
Standard Ethereum smart contract access control offers either full owner permissions or balance-limited permissions with no nuance between those extremes. Zodiac addresses this with a single smart contract using a domain-specific language for defining permission structures, which reduces audit burden and vulnerability surface compared to building bespoke adapters for each protocol. The closest functional parallel is Fireblocks, but Fireblocks operates at the private key level rather than the smart contract level and can be used alongside Zodiac. Zodiac can be deployed on top of any existing Safe without requiring a different base contract, operating through the exec-transaction-from-module entry point so any enabled module can execute transactions within defined constraints.
The Bybit exploit involved a compromised container serving the Safe front end that injected code targeting Bybit specifically and replaced the transaction payload with a malicious one at broadcast time. Macmillan argues Zodiac could have minimized the blast radius because the exploited transaction was a routine fund transfer that could have been pre-defined as a constrained role rather than requiring full root access to the Safe.
Macmillan describes general sentiment around DAOs as currently quite negative, with many observers writing them off as an organizational model. He argues the DAO failure rate may not be worse than the traditional startup failure rate, but failures happen more publicly with thousands of participants involved early on. Early referendum-style governance where everyone votes on everything proved hugely inefficient, and the trend has moved toward hierarchical structures such as sub-DAOs or outsourcing operational work to foundations and labs. Lido, Arbitrum, and Uniswap are cited as examples of large-scale DAOs still making technical, business, and legal progress. Macmillan identifies futarchy and AI agents as two trends that could shift DAOs back toward more broadly distributed decision-making, with AI agents potentially acting as proxies that vote on behalf of token holders based on a corpus of the holder's stated beliefs.
The Interfold, previously called Enclave, emerged from a limitation Macmillan encountered while running ClearFund, a quadratic funding tool built on top of MACI. MACI provides secret ballot voting using zero-knowledge proofs but requires a single trusted coordinator who can view all private state. ZK alone cannot distribute the coordinator role because someone must know all votes to generate proofs of the tally, and ZK cannot achieve collective computation over inputs from multiple parties while keeping those inputs private from one another. The Interfold combines fully homomorphic encryption, zero-knowledge proofs, and MPC to achieve encrypted execution with end-to-end verifiability. ZKPs provide verifiability and client-side input validation without exposing data, FHE allows computation over data while it remains encrypted, and MPC is used specifically to distribute trust over decryption keys. Adding FHE limits the MPC component to only key generation and decryption, so nodes do not need to be online during intermediate computation, which solves the scalability problem that arises in pure MPC systems where every node must participate throughout the entire program execution.
The two primary use cases are voting and sealed bid auctions, which share the property that legitimacy depends on maintaining secret state. A second category involves avoiding liability from touching plaintext data, such as collaborative research and analytics across competing institutions. Macmillan notes that the hospital data use case frequently pitched for ZK systems was technically incomplete when only ZKPs were available, and the FHE-MPC-ZKP combination now fulfills that pitch, though he acknowledges the real challenge for institutional adoption is getting compliance sign-off from competing institutions, which is a human coordination problem more than a technical one. On prediction markets, Macmillan argues the more interesting privacy question is not buyer-side anonymity but privacy over the current state of the market itself, and he illustrates traceability risk with the example of a soldier who bought positions ahead of a mission into Venezuela and is now being tracked through publicly visible on-chain activity.
Each encrypted execution environment in The Interfold spins up for a single job, a distributed key generation is performed for that job, and the key is discarded afterward, making the environments ephemeral by design. The project was renamed from Enclave because people consistently concluded after detailed architecture discussions that it was a network of trusted execution environments, which it is not. Developer documentation and a dev net are already live, with a public-facing testnet expected within weeks of recording and a mid-June production launch on mainnet Ethereum and potentially some L2s targeted following audit work.
This summary was generated from the episode transcript and can contain mistakes.