ProvAuditChain: A Gas-Efficient On-Chain Provenance Framework for AI-Driven Smart Contract Audits

Abstract

The increasing use of AI-powered tools for smart contract security audits presents a critical challenge: ensuring the integrity and provenance of audit reports. Traditional methods lack cryptographic guarantees linking audit outputs to specific AI models and source code, creating vulnerabilities to tampering and misattribution. To address this, we propose ProvAuditChain, a gas-efficient, hybrid on-chain/off-chain framework that records immutable provenance of AI-driven audit reports on Layer 2 blockchain networks. ProvAuditChain utilizes lightweight smart contracts to securely anchor cryptographically signed audit report hashes on-chain, while storing the complete reports on decentralized IPFS storage. We deploy the system on the Arbitrum Sepolia testnet and benchmark gas consumption, latency, and throughput across 600 audit cycles. Our results demonstrate a stable average gas cost of approximately 173,000 per audit, equivalent to roughly $0.05 USD, alongside a throughput of nearly six audits per minute. These findings confirm the practical viability of ProvAuditChain for integration into automated CI/CD pipelines, providing a scalable foundation for trustworthy AI accountability in decentralized ecosystems.