Why use collaborative SNARKs instead of just MPC? We often get asked: if you already trust an MPC network for privacy, why put ZK on top? If there’s a trust assumption anyway, why not also trust it for correctness? 🧵

1/ ZK is becoming the default way to verify compute. First onchain, then probably everywhere. If your output only relies on trust in the MPC network, people will start to avoid it.

2/ you might not trust the MPC network with your own data, but others do. If the network proves its computation via ZK, you can still trust the result—even if you don’t trust the network that produced it. Now, why not also remove the trust assumption for privacy?

3/ I wish we could. But right now, there’s no practical way. That said, more and more people are working on reducing or even removing this assumption (including us). If we make the same kind of progress here as we’ve seen in ZK, we might actually get close to a ideal solution for private shared state.