PQ-AgentID: Post-Quantum, Stateless-Verifiable Identity for Autonomous Agents on Ethereum

Autonomous AI agents are increasingly expected to hold wallets, sign transactions, interact with other agents, and accumulate long-lived on-chain reputations. Standards such as ERC-8004 provide an emerging foundation for agent identity, reputation, and validation on Ethereum, but their reliance on classical elliptic-curve signatures creates a long-term security concern in the presence of cryptographically relevant quantum computers. This paper identifies and formalizes a harvest-now, forge-later threat for on-chain agent identities. Unlike the commonly discussed harvest-now, decrypt-later threat, the concern here is that an adversary can record publicly available identities and signatures today and potentially forge future actions under those identities once the underlying classical signature scheme becomes vulnerable to quantum attacks. The paper proposes PQ-AgentID, a hybrid architecture for post-quantum, stateless-verifiable identity for autonomous agents on Ethereum. The architecture separates an agent's long-lived root identity from its high-frequency operational activity. The root identity is protected using the NIST-standardized ML-DSA-44 post-quantum signature scheme, while short-lived session keys are used for frequent agent interactions. A Verkle-tree-based delegation commitment structure binds session keys to the post-quantum root identity and allows batches of delegations to be committed on-chain while remaining independently verifiable off-chain. The paper provides four main contributions. First, it formalizes the harvest-now, forge-later threat model for on-chain agent identity registries. Second, it proposes the PQ-AgentID hybrid identity and delegation architecture. Third, it provides an explicit adversary and trust model together with a security analysis. Fourth, it develops a first-principles analytical cost model comparing the classical ERC-8004 baseline, naive post-quantum signature substitution, and the proposed batched construction. Using published NIST parameter sizes, the analysis shows that directly replacing 64-byte classical signatures with 2,420-byte ML-DSA-44 signatures increases signature data by approximately 37.8×. PQ-AgentID instead amortizes post-quantum signatures across delegation batches, allowing the post-quantum component of the cost to grow with the number of batches rather than with every agent interaction. The work is an architectural and analytical research contribution. It does not claim a completed production implementation, empirical gas benchmark, latency benchmark, or formal machine-checked protocol proof. Implementation, empirical benchmarking, root-identity recovery mechanisms, and post-quantum vector-commitment extensions are identified as directions for future work.