Soda Labs co-founder Avishay Yanai on garbled circuits, the four-party disclosure stack, and why “MPC vs. FHE” is the wrong question.
TL;DR
→ Soda Labs co-founder and CEO Avishay Yanai makes the case that real privacy infrastructure is operator-defined: the business that owns the smart contract decides what is visible, and to whom.
→ Institutional privacy is a layered problem across four parties: the regulator, the AML/KYC partner, the real-time security provider, and the counterparty. The architecture has to serve all four.
→ His case for Garbled Circuits rests on four properties: integration, flexibility, throughput, and deployment economics.
→ It is the same cryptography, developed with Soda Labs, that has powered COTI’s mainnet since March 2025.
Who gets to decide what stays private?
That question sits under every serious conversation about privacy on public blockchains. In a new interview with Proof Street, Soda Labs co-founder and CEO Avishay Yanai gives an answer that reframes the debate: the operator decides.
Yanai is one of the cryptographers behind gcVM, the garbled-circuit virtual machine COTI’s network runs on. So when he describes what privacy infrastructure has to do, he is describing the same technology that has been in production on COTI since March 2025. Here is why his framing matters, and how it maps onto what COTI has already shipped.
Privacy Is a Decision, Not a Default
Yanai’s core claim is simple. The business that owns a smart contract should decide what privacy looks like, and who is allowed to see what. If a user accepts the service, they accept its terms, the same way they do with any service online.
What changes on-chain is verifiability. In an ordinary service, you trust that a provider handles your data the way its terms promise, with no way to check. When privacy is a property of the cryptography itself, the promise and the behavior become the same thing. Anyone can verify who can see what, and when.
That is the core of Privacy-on-Demand. Privacy is not a fixed setting bolted onto the chain. It is programmable, configured by the operator to fit the use case in front of them.
The Four-Party Stack
Yanai’s second point: institutional privacy is never a single switch. In most transactions with privacy implications, four parties are looking at the data, each with a different level of access and a different reason to be there.
The regulator. For anyone handling money or regulated data, disclosure to regulators is mandatory. The privacy posture has to stay compatible with a subpoena.
The AML and KYC partner. External providers contracted to monitor for illicit activity. They get a defined level of visibility: more than nothing, less than full plaintext.
The real-time security provider. The least-discussed layer. Firms that watch smart contract activity as it happens and can block a suspicious transaction before it executes, often by acting as a required co-signer.
The counterparty. The layer most user-facing privacy conversations focus on. In institutional settings it sits at the bottom of the stack, seeing only what the operator agreed to share.
The real question for any architecture, Yanai argues, is whether it lets the operator define a clean posture across all four layers, or forces a single posture the technology dictates. Good infrastructure has to be operator-defined from the ground up.
This is the model COTI describes as compliant confidentiality: selective disclosure, where an institution reveals exactly what it needs to, to exactly whom, and nothing more.
MPC Is a Problem, Not a Solution
Here Yanai reframes the category. Multi-party computation (MPC) is usually pitched as a rival to FHE, ZK proofs, and TEEs. He pushes back: MPC names a problem, compute on data without revealing the inputs, not one technique that solves it.
His sharpest example is the FHE comparison. Production FHE systems for blockchain lean on threshold MPC for key management. Several leading FHE projects, Yanai notes, are MPC protocols underneath, with FHE handling one specific operation. The popular “FHE versus MPC” split is less clean than the marketing suggests.
That reshapes how builders should choose. The right question is not “which is fastest.” It is: what disclosure posture does my use case need, and which technique supports it at the throughput I require?
The reframe matters for COTI, because it moves the conversation past raw benchmarks and onto architecture, where the real differences decide what you can actually build.
Why Garbled Circuits
So why build on Garbled Circuits, a cryptographic protocol for secure computation that has existed since the 1980s? Yanai gives four reasons. In plain terms: it runs almost anywhere, it fits how apps already work, it is fast enough for real use, and it is cheap to operate.
Integration: it runs on encryption your devices already trust. Garbled Circuits, as Soda builds them, use AES, the same standard encryption that has protected everyday internet traffic for more than two decades. Because it relies on tools already built into phones, laptops, cloud servers, and even small connected devices, there is nothing exotic to install. The privacy layer speaks a language almost every device already knows, so it can go almost anywhere.
Flexibility: it fits how apps and tokens already work. Developers can use it without redesigning their applications, because it handles the same kind of token math standard Ethereum apps already rely on. It can also do something subtle but useful: let someone prove they hold a token without revealing which specific one. Yanai’s example is a trading firm. Its orders still show up in the market, where they need to be, while the link between a trade and the firm behind it stays confidential.
Throughput: it is fast enough for real-time use. The trick is timing. The slow, heavy math is done in advance, in the background, so the live step is quick when a real transaction arrives. Picture a kitchen prepping ingredients before the dinner rush, so plates go out fast once orders land. Soda’s measurements put this at 83 confidential transactions per second today, with a path to 400 to 500 as the behind-the-scenes work is optimized.
Deployment economics: it is cheap to run at scale. It works on ordinary computers. There is no need for the specialized, expensive, hard-to-source hardware that some other privacy approaches depend on. That keeps costs low, which matters most as a network grows to many operators, each running their own machines.
Yanai’s point is that fast was never the whole argument. The reason to choose this cryptography is that it lets the operator define privacy across all four layers, and lets anyone outside that perimeter verify the computation ran faithfully without seeing any of the underlying values. That property, public auditability, is the real differentiator.
How COTI Fits In
None of this is theoretical for COTI. It is live.
The same gcEVM cryptography has powered COTI’s mainnet since March 2025, the first on-chain implementation of Garbled Circuits in production. COTI gcEVM has powered real end-to-end encryption use cases such as 10-million vaccine ledger transactions, private RWAs with Zoniqx, and now private tokens with the COTI privacy portal.
COTI pairs this with a dual-privacy stack: COTI GC for fast, programmable confidentiality, and COTI Nightfall, an enterprise-grade Zero-Knowledge rollup for regulated institutional workflows. Two techniques, one operator-defined posture, chosen per use case. That is the flexibility Yanai argues privacy infrastructure has to have.
The Bigger Picture
Yanai also points to where the technology is heading. Soda’s own product line now spans two shapes: the integrated approach COTI runs on, where privacy is a property of the chain, and a composable coprocessor an operator can add to an existing stack without migrating chains. Different form factors, same core idea: privacy the operator composes in, on their terms.
Asked where he wants it to go, Yanai describes a large network of MPC nodes serving most of the financial activity on blockchains. His reasoning is pragmatic. Blockchains offer repeatable, structured use cases, token transfers, trades, settlement, so infrastructure can be built once and consumed many times. That repeatability, not the cryptography alone, is what makes the business scale.
It reframes the question institutions will ask. Not “which privacy chain do I migrate to,” but “which privacy layer do I compose into the stack I already run.” That is the Privacy-on-Demand thesis, described from the cryptographer’s seat.
Privacy that is selective, layered, programmable, and defined by the business. That is what the market is asking for. It is what COTI is already delivering.
This piece adapts and paraphrases an interview by Stephen Webber for Proof Street. Read the original: Defining Privacy in MPC from the Operator’s Seat.
Stay COTI.
About COTI:
COTI is the programmable privacy layer for Web3. Built for enterprises, builders, and agents. Powered by high-performance Garbled Circuits and enterprise-grade COTI Nightfall (ZK), COTI enables encrypted computation on any public blockchain. Fast, low-cost, and compliant privacy across DeFi, AI, and beyond.
For COTI updates and to join the conversation, be sure to check out our channels:
Website: https://coti.io/
X: https://twitter.com/COTInetwork
YouTube: https://www.youtube.com/channel/UCl-2YzhaPnouvBtotKuM4DA
Telegram: https://t.me/COTInetwork
Discord: https://discord.gg/coti-foundation
GitHub: https://github.com/coti-io
Vibe Coders Telegram: https://t.me/+uuPNfRkKiQ03ZTcx