Is crypto needed to protect the security of AI agents paying each other online?
The security of AI agents paying each other hinges on solving the 'who decides the work was done' verification problem, which crypto standards like ERC-8183 attempt to address via programmable escrow.Summary
The burgeoning field of agentic commerce involves numerous protocols standardizing how AI agents connect, coordinate, and initiate payments (e.g., Anthropic's MCP, Google's UCP, Coinbase's x402). However, these standards largely ignore the critical question of verification: who judges when a task is satisfactorily completed? Incumbents like Mastercard and Google are focusing on authorization and audit trails (Verifiable Intent, Agent Payment Protocol), proving a purchase was sanctioned, but not that the outcome materialized. Crypto's primary attempt to solve this missing primitive is ERC-8183, a draft Ethereum standard defining a job-based escrow flow where funds are locked until an evaluator attests to completion. Critics note this is essentially programmable escrow, not inherently 'agentic.' The core tension lies in who controls the 'evaluator' role, which holds platform-level leverage in multi-party agent networks. The future of agentic commerce security depends on whether Big Tech's authorization standards or on-chain programmable escrow, like ERC-8183, ultimately owns this crucial judgment layer.
(Source：CryptoSlate)