DAO-Agent: Zero Knowledge-Verified Incentives for Decentralized Multi-Agent Coordination

Yihan Xia, Taotao Wang, Wenxin Xu, Shengli ZhangarXiv 2025 · OpenReview

TL;DR

This paper introduces DAO-Agent, a novel framework that enables decentralized multi-agent systems (MAS) to coordinate and distribute incentives fairly and securely, even in trustless environments. By combining blockchain, Zero-Knowledge Proofs (ZKPs), and Shapley values, DAO-Agent drastically reduces on-chain costs while ensuring verifiable and private contribution measurement.

What Problem Does This Paper Address?

Autonomous Large Language Model (LLM)-based multi-agent systems are becoming increasingly powerful for complex collaborations. However, coordinating these agents in trustless environments presents significant challenges:

  • Transparent Contribution Measurement: How do we accurately measure each agent's contribution when their outputs are interdependent?
  • Equitable Incentive Distribution: How can we ensure fair rewards that incentivize honest participation without exposing private strategies?
  • Scalability and Cost: Traditional blockchain solutions for coordination are computationally expensive, making them impractical for complex MAS.
  • Trustlessness: Centralized coordination mechanisms introduce single points of failure and trust issues.

Existing solutions struggle to balance decentralization, transparency, fairness, and computational efficiency. This paper tackles the dilemma between "Code is Law" (maximal transparency but high cost) and off-chain systems (abundant compute but lack of verifiable trust).

Methods

DAO-Agent proposes a "Compute Off-chain, Verify On-chain" paradigm, integrating three key technical innovations:

  1. On-chain DAO Governance: A decentralized autonomous organization (DAO) mechanism for transparent coordination and immutable logging of task progress.
  2. ZKP-based Shapley Value Measurement: A Zero-Knowledge Proof (ZKP) approach to measure agent contributions off-chain using Shapley values. Shapley values are a game-theoretic concept that axiomatically quantifies an agent's marginal contribution to a coalition, ensuring fairness.
  3. Hybrid On-chain/Off-chain Architecture: A system that separates computationally intensive Shapley value calculations (off-chain) from lightweight on-chain verification. This is achieved through a recursive ZKP pipeline (STARK proofs compressed into Groth16 proofs) that allows the blockchain to verify the fairness of the off-chain computation with minimal computational overhead.

The framework uses a dual-commitment scheme: agent outputs are hashed and stored on IPFS with a Content Identifier (CID) anchored on-chain, and coalition values are also hashed and committed. This ensures output authenticity and tamper-proofing.

Datasets

Not discussed in the paper. The paper uses a simulated cryptocurrency trading task as a case study for experimental evaluation.

Key Insights and Ideas

  • Shapley Values for Fair Incentives: The paper highlights how Shapley values provide a mathematically sound and axiomatically fair way to distribute rewards in collaborative settings, making honest participation a dominant strategy for rational agents.
  • ZKPs as the "Justice" Layer: By using ZKPs, DAO-Agent replaces reliance on trusted intermediaries with mathematical proofs, transforming "trust in intermediaries" into "trust in mathematics." This ensures verifiable fairness without revealing private strategies.
  • Hybrid Architecture is Key to Scalability: The core innovation is decoupling expensive computations (Shapley value calculation) from cheap verification. This allows complex MAS to leverage blockchain's trust properties without being crippled by its computational limitations.
  • Constant-Cost Verification: The ZKP pipeline compresses complex off-chain computations into a constant-size proof, enabling on-chain verification with a fixed cost (around 27k gas) regardless of the number of agents. This is a significant departure from exponential on-chain costs.

Key Results

  • Massive Cost Reduction: DAO-Agent achieves up to a 99.9% reduction in verification gas costs compared to naive on-chain Shapley value calculation for a 10-agent system.
  • Constant Verification Complexity: The on-chain verification cost remains stable at approximately 27k gas, independent of the number of agents.
  • Scalable Verification Time: On-chain verification time remains negligible and constant (under 0.4 seconds) even as the number of agents increases.
  • Proof Size Management: While off-chain STARK proof generation scales exponentially, the subsequent compression to Groth16 proofs results in small, manageable proof sizes that increase only marginally with agent count.

Limitations and Future Work

  • Off-Chain Proof Generation Bottleneck: The primary scalability constraint for the off-chain process is the exponential time required for STARK proof generation, which scales with the number of agents (O(2^n)).
  • Coordinator Liveness Assumption: The current security model assumes the off-chain coordinator is "live" (i.e., it will eventually perform its tasks) but not necessarily correct, with correctness enforced by ZKPs.
  • Future Work Suggestions:
    • Approximation Techniques: For even larger-scale deployments, Monte Carlo Shapley estimation could be used to address the exponential proof generation time.
  • Enhanced Security: Redundant coordinators or decentralized challenge-response mechanisms could strengthen the system's security model.
  • Post-Quantum Cryptography: Adapting the framework to post-quantum hash functions for future-proofing.
  • Layer-2 Deployment and Proof Batching: Exploring these techniques to further enhance performance.

Summary updated

BibTeX
@misc{xia2025dao,
  author = {Yihan Xia and Taotao Wang and Wenxin Xu and Shengli Zhang},
  title = {{DAO-Agent: Zero Knowledge-Verified Incentives for Decentralized Multi-Agent Coordination}},
  year = {2025},
  url = {https://openreview.net/forum?id=JUYtJ51lgU},
}
Figure