The new protocol, developed with founding contributors including Google, IBM, Circle, Wayfair and UiPath, creates a verifiable record of what a shopper authorized and what an AI agent decided on its own, addressing a gap that has left many consumers uncertain about the extent of their consent when agents act on their behalf.

Under the standard, merchants publish their terms at a fixed web address; an AI agent can prove which version it saw, and both parties sign off on the agreement. The system does not require blockchain or a third‑party intermediary, allowing any organization with a web server to adopt it.

The protocol splits a transaction into two sides: the consumer‑side captured by tools such as Google’s Agent Payments Protocol, and the merchant‑side captured by the Legal Context Protocol, which logs the offer, accepted obligations and any recourse mechanisms. Four levels of proof are defined, ranging from simple consent to a digital signature that can trigger arbitration, escrow or compliance processes.

A key feature is the “buyer policy,” which lets shoppers set spending caps, jurisdiction limits, preferred dispute bodies and thresholds that trigger human review. When a purchase exceeds a set amount, involves a first‑time merchant, or falls under an unfamiliar jurisdiction, the agent’s signing path halts until a person or supervising agent approves the transaction.

Consumer trust in AI‑driven payments remains low. A PYMNTS Global Digital Shopping Index survey of 5,241 consumers found that only 23% of U.S. shoppers trust generative AI to handle payment transactions, and trust drops further as the agent approaches the final payment step.

Retailers are already adjusting their terms to accommodate agents. Target, for example,

Industry forecasts suggest rapid growth in agent‑mediated commerce. Bloomberg Intelligence projects that by 2030 about 20% of online commerce—roughly $500 billion—will be conducted through agents acting on a single prompt, up from less than 5% this year, with an additional 6.2% of transactions expected to be fully autonomous.