Is crypto needed to protect the security of AI agents paying each other online?

The burgeoning field of agentic commerce is witnessing a rapid development of infrastructure protocols, yet a critical gap remains: verifying that completed work meets agreed-upon standards. While major players like Anthropic, Google, and Coinbase are establishing standards for how AI agents connect, coordinate, and initiate payments, the harder question of who decides when work is truly done is largely unaddressed by these initial efforts.

Existing Protocols Address Connectivity and Payments

Several protocols are emerging to standardize different aspects of agentic commerce. Anthropic's Model Context Protocol (MCP) connects AI applications to external tools and data, seeing significant adoption with over 10,000 public servers and 97 million monthly SDK downloads. Google's Agent-to-Agent protocol (A2A), now under Linux Foundation governance, facilitates communication and coordination across systems. The Universal Commerce Protocol (UCP), launched by Google with support from major retailers like Shopify, Walmart, and Target, along with payment giants like Mastercard, Visa, and American Express, aims to standardize agent-driven checkout flows. Coinbase's x402 protocol handles the payment transport layer, enabling automatic stablecoin payments over HTTP, having processed over 100 million payments by late 2025.

These protocols, however, primarily focus on the mechanics of connection and payment authorization. They do not solve the crucial problem of verifying the quality or completion of the work performed by agents. For instance, MCP handles the tool/data layer but not the trust around completed work, A2A solves agent interoperability but not conditional settlement, and UCP pushes agents into transactions without verifying satisfactory task completion. Similarly, Google's Agent Payment Protocol (AP2) and Coinbase's x402 focus on authorization and payment transport, respectively, not on the verification of delivered outcomes.

ERC-8183 Proposes a Solution for Conditional Settlement

The draft Ethereum standard ERC-8183, published on February 25, 2026, attempts to address this verification gap by introducing a programmable escrow primitive for task-based commerce. The proposal outlines a minimal state machine where a client locks funds into escrow, a provider submits work, and an evaluator marks the job as complete or rejected. The system includes automatic refunds upon expiry. While framed within agentic commerce, critics note its core functionality as a job registry with escrowed funds, applicable to any task-based transaction, human or machine.

ERC-8183 aims to provide the missing conditional payment and verification step in the agentic commerce stack. It is seen as a potential composition layer, especially when combined with protocols like ERC-8004, which provides a trust and reputation framework for agents. However, ERC-8183 itself does not solve evaluator trust or dispute resolution, leaving the critical role of the evaluator as a potential bottleneck.

The Authorization-Verification Gap and Incumbent Involvement

Incumbent payment providers are also developing solutions that touch upon agentic commerce. Google's Agent Payment Protocol and Mastercard's Verifiable Intent, co-developed with Google, focus on creating trust and audit layers for purchase authorization. These solutions address whether a purchase was sanctioned but do not verify if the purchased outcome was delivered. This highlights a divergence where established players focus on authorization, while crypto-native solutions like ERC-8183 target the verification and conditional settlement layer.

The Power of the Judgment Layer

The evaluator role within ERC-8183 is politically significant. The proposal acknowledges that a malicious evaluator can arbitrarily complete or reject jobs, recommending reputation or staking mechanisms for higher-value contracts. The core spec lacks dispute resolution, leading to observations that "