Secure Verification of AI-Generated Text Using a Blockchain-Based Dynamic Watermarking Scheme with Zero-Knowledge Proofs

In this paper, we propose a dynamic watermarking method for LLMs that utilizes blockchain technology and zero-knowledge proofs (ZKPs) to provide verifiable authenticity and integrity of digital assets. Our approach makes use of a smart contract deployed on a blockchain network to generate unique, tamper-resistant signatures based on a vendor's secret key and a time-dependent parameter. To ensure the validity of these signatures without revealing the underlying secret information, we design a zero-knowledge proof circuit in Circom using the Poseidon hash function. The proof is then verified on-chain via a Groth۱۶-based verifier smart contract. Our implementation and experimental evaluation show that the average signature generation takes ۰.۰۶۸۷ seconds, proof generation requires ۳.۶۴ seconds, and on-chain verification consumes approximately ۴.۹۲ million gas. The proof size is compact, averaging ۸۰۴ bytes. These results demonstrate the feasibility of our scheme for real-world scenarios where privacy, authenticity, and efficiency are critical.

کلیدواژه ها:

Dynamic watermarking ، zero-knowledge proof ، blockchain ، Groth۱۶ ، Circom