ZK VMs made verifiable computation accessible to any developer. The prover networks running them require your full plaintext data. YIKES

Zero-knowledge cryptography went through three phases. First: hand-crafted arithmetic circuits, only accessible to deep researchers. Second: ZK virtual machines — suddenly any developer could write verifiable code in Rust or C. Third: prover networks (Succinct, Boundless/RiscZero) that let you delegate the heavy proof generation to external infrastructure.

Each phase made the technology more accessible. Each phase also moved the user's data further from their control.

Prover networks require your full plaintext data to generate proofs. For rollups, this is a non-issue — public ledger, no privacy expectation, and what you gain (succinctness — compressing thousands of transactions into a single proof) is worth the trade. That's the use case these networks were built for, and they served it well.

The problem emerges when you extend this model to user-facing applications. Verifiable identity: proving you hold a valid passport, proving you're over 18, without disclosing the underlying data. Private AI inference: running a model on your data without the model owner seeing your inputs or you seeing their weights. Decentralized exchanges with private order books. In all of these, delegating to a prover network means surrendering exactly the inputs you need to keep private.

I sat down with a researcher at ChainSafe who's working on this specific problem. His approach: adding MPC (multi-party computation) to ZK VMs so proof generation can be delegated privately. Multiple parties each hold a secret share of the data, compute their portion, and combine results — no single party ever sees the full picture. He calls it "make ZK VMs ZK again."

He also covered a near-term approach to the deepfake problem: attested sensors that cryptographically sign photo/video metadata at capture, combined with verifiable edit histories. You can't yet verify what IS AI-generated. But you can prove everything that is human — a reverse approach. Prove provenance instead of detecting fakes.

The full conversation covers ZK, MPC, and FHE (the "holy trinity of programmable cryptography"), explained through photography analogies that are genuinely useful for building intuition. We filmed it across Taipei — street markets, a botanical garden, a tea ceremony.

Full interview: https://youtu.be/PnEivfTpnA8

—————

If we're meeting for the first time, hi ????! I started building my channel to spread the good word on good work in crypto — something with substance and humanity. A like, sub, and comment goes a long way to supporting me, so please consider doing so!