Skip to content
LOOKOUT
← LandscapeSector

Verification & Trust.

The trust layer for machine intelligence — proving an AI did what it claims, on-chain, without revealing the model.

Tracked 07Conviction 03Watching 03Skeptical 01

Thesis

Why this sector, why now.

As agents transact and models earn on their own, the unanswered question is verification: did the inference actually run, on the model claimed, producing the output that got paid for? zkML, optimistic ML, and proof-of-inference turn "trust me" into "verify me." This is the bucket where Lookout's CertiK relationship is a genuine edge — auditing and proof systems are adjacent disciplines, and no neutral platform covers verifiable AI with audit-grade credibility. The fork: do proofs live at the model layer (zkML — EZKL, Modulus, Giza) or the coprocessor layer around it (Lagrange, Space and Time)? My read: zkML is still too expensive for large models, so near-term value accrues to optimistic and coprocessor approaches; zkML wins later as proving costs fall.

Signals I track

What would move my read.

  1. 01

    Proving cost per inference vs model size (the zkML scaling wall)

  2. 02

    Real protocols consuming proofs in production, not benchmarks

  3. 03

    Whether audit / insurance products attach to proof layers (the CertiK-adjacent wedge)

Kill shot

What would kill the thesis

Trusted execution environments (Intel SGX, Nvidia confidential compute) get cheap and 'good enough' for verifiable inference, and the market never pays the premium for cryptographic proofs.

Coverage

Projects on the radar.

Lagrange

Conviction

ZK coprocessor and decentralized prover network for verifiable big-data SQL and AI inference; $13.2M seed led by Peter Thiel's Founders Fund.

Deep read →

RISC Zero

Conviction

Seattle team behind the leading general-purpose RISC-V zkVM, Bonsai proving service, and the Boundless (ZKC) proof marketplace; ~$52M from Bain Capital Crypto and Blockchain Capital.

Deep read →

Space and Time

Conviction

Microsoft-backed ZK-proven (Proof of SQL) verifiable data warehouse and L1 for trustless data + AI; $40M total incl. M12 and Framework.

Deep read →

EZKL

Watching

Developer-first zkSNARK engine that proves ONNX machine-learning inference ran correctly; a small but widely-used ~$3.8M-backed primitive of the zkML stack.

Deep read →

Giza

Watching

Verifiable-ML protocol that pivoted from zkML tooling into autonomous DeFi agents (ARMA); ~$8.2M from CoinFund and Coinbase Ventures, GIZA token live since May 2025.

Deep read →

Ora Protocol

Watching

Onchain AI oracle tokenizing models via opML + Initial Model Offerings; $20M from Polychain, HF0 and HashKey.

Deep read →

Modulus Labs

Skeptical

Stanford-founded zkML prover ('bring AI on-chain') that raised $6.3M from Variant and 1kx, then wound down as an independent entity — acquired by Worldcoin's Tools for Humanity in Dec 2024.

Deep read →

Going deeper

Bespoke Verification & Trust dive for your fund.

Brief me

Draft thesis · editorial voice in progress, edits land continuously