Skip to content
Crypto.world

Research

zk-Agreements: A Privacy-Preserving Way to Establish Deterministic Trust in Confidential Agreements

Paper contracts and even digitized "computable legal contracts" are slow to enforce and leave dispute resolution to courts, while running those contracts as smart contracts on a public blockchain…

zk-Agreements: A Privacy-Preserving Way to Establish Deterministic Trust in Confidential Agreements
arXiv
2510.20007
Published
2025-10-22
Authors
To-Wen Liu, Matthew Green

AI summary

Overview

Research area: Applied cryptography and privacy-preserving blockchain systems, at the intersection of zero-knowledge proofs, trusted execution environments, secure multi-party computation, and computable legal contracts (cs.CR).

Technical level: Advanced. The paper assumes familiarity with zk-SNARKs, commitment schemes, Merkle trees, game-based security definitions, and trusted hardware, though the motivating narrative about contracts and dispute resolution is accessible to non-specialists.

Scope: The paper proposes and benchmarks zk-agreements, a protocol that keeps confidential contract terms off a public ledger while still using a blockchain and zero-knowledge proofs to enforce settlement deterministically, illustrated with a rental security deposit case study.

What This Paper Is About

Paper contracts and even digitized "computable legal contracts" are slow to enforce and leave dispute resolution to courts, while running those contracts as smart contracts on a public blockchain like Ethereum exposes confidential business terms to anyone. zk-agreements aims to get both properties at once: the automated, deterministic enforcement of a blockchain without publishing the agreement's contents. The authors combine zero-knowledge proofs (PLONK over BN254), evaluation inside a Trusted Execution Environment, and an on-chain smart contract verifier so that only a commitment, a compliance ratio, and a succinct proof appear publicly.

Key Contributions

  1. A privacy-preserving legal execution architecture. A hybrid design combining zk-SNARKs, secure multi-party computation, and smart contracts that preserves contractual confidentiality while retaining legal enforceability; the authors state this is the first protocol to combine legal enforceability with cryptographic privacy for confidential business agreements.
  2. Verifiable off-chain contract evaluation. A protocol in which independent evaluators collect real-world data inputs and forward them to a TEE, which checks contractual terms and produces succinct proofs used for on-chain settlement, without revealing the underlying business data.
  3. A legal-cryptographic bridge. Automated translation of natural-language agreements into cryptographically enforceable artifacts using the Accord Project toolchain (Ergo for contract logic, Concerto for data models, Cicero as a template engine).
  4. A six-algorithm protocol specification plus a formal security analysis. SignCLC, CommitCLC, SubmitCommitment, ExecuteEvaluation, GenerateProof, and VerifyProof, with game-based arguments for privacy, soundness, non-repudiation, and TEE correctness/confidentiality, backed by circuit benchmarks.

Main Findings

  • Confidentiality rests on three stacked assumptions. The privacy argument reduces to (i) hiding of the contract commitment, (ii) zero-knowledgeness of PLONK, and (iii) TEE isolation preventing leakage of intermediate computation state; the adversary's distinguishing advantage is required to be negligible in the security parameter.
  • Soundness holds if one party is honest and the TEE runs correctly. The paper states correct evaluation is output with probability at least 1 − 2^−λ, relying on remote attestation, PLONK soundness, and blockchain immutability.
  • Non-repudiation depends on signature unforgeability and ledger immutability. Signatures over the commitment clc_comm = Hash(k ‖ pd) are recorded on-chain, and a nullifier table prevents reuse of an execution.
  • Settlement is proportional, not binary. On a valid proof, the seller receives v × rat and the buyer receives v × (1 − rat), where v is the total contract value and rat the compliance ratio produced inside the TEE.
  • Circuit costs favor verification over proving. On an Apple M2 Pro laptop (10-core CPU at 3.5 GHz, 16 GB RAM), the commitment circuit compiles in 520 ms, generates a witness in 70 ms, runs setup in 260 ms, proves in 260 ms, and verifies in 270 ms; the evaluation circuit takes 110 ms, 30 ms, 670 ms, 2,180 ms, and 260 ms respectively.
  • The evaluation circuit is small but expensive to prove. It has 1,781 wires, 1,499 constraints, 277 private inputs, 5 public inputs, 2,413 labels, and 0 outputs; the commitment circuit has 1,870 wires, 1,357 constraints, 512 private inputs, 0 public inputs, 24,130 labels, and 2 outputs.
  • On-chain cost comparison. Table 1 lists approximate gas of 1.2M for a commitment and 0.35M for an evaluation, versus approximately 21K for an Ethereum transfer, approximately 50K for an ERC-20 transfer, approximately 7.19M for Zether, approximately 1.1M deposit and approximately 0.4M withdrawal for Tornado Cash; OpenLaw is listed as N/A. zk-agreements and OpenLaw are the only entries marked as providing legal enforceability.
  • Full end-to-end deployment metrics are not reported. The paper benchmarks the Circom circuits only; no mainnet deployment gas measurements, latency for the TEE evaluation path, or throughput figures appear in the provided content, and Appendix 0.B (the detailed case-study specification) is truncated where the provided text ends.

Methodology in Plain English

The parties start by agreeing on a contract in ordinary legal text and convert it into machine-readable form with Accord Project tools: Concerto describes the data (who the parties are, amounts, dates), and Ergo encodes the clauses as executable logic.

Both parties sign a hash of that contract. A secret nullifier is combined with a digest of the signed contract to produce a single commitment value, which is posted to a smart contract on-chain and stored in a Merkle tree. The blockchain now sees only that opaque commitment and the total value at stake, not the terms themselves. The contract moves through four states: INIT, EXECUTION, EVALUATION, and COMPLETED.

When performance has to be judged, an evaluator feeds real-world evidence (for example, a property inspection report) into a TEE along with the signed contract. Inside the sealed hardware, the contract logic runs and produces a compliance ratio. The TEE then produces a zero-knowledge proof that the evaluation was done correctly against a contract whose commitment is genuinely in the on-chain Merkle tree, without revealing the contract or the evidence.

The smart contract checks the proof is valid, checks the nullifier has not been used before, and splits the escrowed funds according to the compliance ratio. The rental deposit example walks through this: a tenant commits deposit terms privately, the landlord submits inspection evidence at lease end, the TEE evaluates it, and the deposit is released to tenant or landlord automatically.

Why This Matters

Research impact. The paper targets a gap the authors identify explicitly: privacy-preserving blockchain systems such as Zether, Hawk, ZEXE, and Tornado Cash generally do not attempt legal enforceability, while smart legal contract projects such as OpenLaw and Ethereum smart contracts offer little or no confidentiality. zk-agreements is positioned as the first system to claim both, and contributes a concrete protocol template (commitment plus TEE evaluation plus succinct proof plus proportional payout) that others can build on.

Real-world applications:

  • Rental security deposits. The paper's worked example: tenant and landlord escrow a deposit and release it based on a TEE-evaluated inspection report without publishing lease terms or tenant identity on-chain.
  • Supply-chain payment-on-delivery. The Accord Project sample contract (goods costing $10.00 plus a $2.00 delivery fee, $12.00 total due on delivery and acceptance) is presented as encodable in Ergo and Concerto and thus expressible in the protocol.
  • Confidential B2B procurement and service agreements. Commercial terms that would reveal competitive intelligence can be committed rather than published.
  • Automated, low-cost dispute resolution for weaker parties. The paper frames "David versus Goliath" disputes, where a small buyer cannot afford litigation against a large merchant, as a motivating case for deterministic settlement.

Industry relevance. The design keeps the expensive work (contract execution and evidence handling) off-chain in a TEE and puts only a proof verification and a payout on-chain, which is the pattern enterprises typically need before putting negotiated commercial terms anywhere near a public ledger. The stated on-chain costs (approximately 1.2M gas for a commitment, approximately 0.35M for an evaluation) indicate the approach is plausible on Ethereum-class chains but not cheap relative to a plain transfer.

Future Directions

  • Trust assumptions around the TEE. The threat model explicitly excludes physical tampering and side-channel leakage, and assumes attestation can be verified by other parties; hardening or replacing the TEE (for example with multi-party computation alone or threshold attestation) is a natural open question.
  • Real deployment measurement. The paper reports circuit benchmarks on a laptop and approximate gas figures in a comparison table, but does not report measured on-chain gas, end-to-end latency including TEE evaluation, or cost at scale; those measurements are the obvious next step.
  • Handling contract types the evaluation circuit does not cover. The evaluation circuit is only 1,499 constraints with 5 public inputs, suggesting a narrow compliance predicate; extending the approach to richer legal logic and to disputes where evidence itself is contested remains open.
  • Adversarial evaluation and data authenticity. The protocol relies on independent evaluators supplying measured data; how to bind those inputs to reality and punish a colluding or faulty evaluator is not resolved in the provided content.

Target Audience

Cryptography and security researchers working on privacy-preserving smart contracts and zk-SNARK applications; legal-tech and regtech engineers evaluating whether confidential commercial agreements can be automated on a public ledger; and enterprise architects and policy-minded readers who need to understand the tradeoff between on-chain transparency and contractual confidentiality. Readers without a cryptography background can follow the motivation and the rental-deposit example, but the protocol and security sections require comfort with commitment schemes, Merkle proofs, and game-based proofs.

Authors’ abstract

Digital transactions currently exceed trillions of dollars annually, yet traditional paper-based agreements remain a bottleneck for automation, enforceability, and dispute resolution. Natural language contracts introduce ambiguity, require manual processing, and lack computational verifiability, all of which hinder efficient digital commerce. Computable legal contracts, expressed in machine-readable formats, offer a potential solution by enabling automated execution and…

The full abstract on arXiv
Read the original paper