Legal Context Protocol Logs the Terms AI Agents Accept

A consumer who tells an artificial intelligence (AI) agent to book a vacation for under $3,000 has approved a budget, not every choice the agent makes to stay inside it. The agent can pick the airline, accept a nonrefundable fare, add travel insurance and split the trip across several charges without asking again. When something goes wrong, the record of what the shopper allowed and what the agent decided alone often doesn’t exist.

The American Arbitration Association and Integra Ledger launched the Legal Context Protocol in June to build that record. Google, IBM, Circle, Wayfair and UiPath joined as founding contributors alongside more than a dozen blockchain and identity firms. The open standard lets a merchant publish its terms at a fixed web address, lets an agent prove which version it saw and lets both sides sign what was agreed.

“The agentic economy needs that same capacity delivered at machine speed,” Bridget McCormack, president and CEO of the AAA, said in the announcement. The protocol needs no blockchain or intermediary, and any organization with a web server can adopt it.

Budget Approval Is Not Blanket Consent

The protocol splits a transaction into two sides. Authorization tools such as Google’s Agent Payments Protocol capture the consumer’s side, and the new standard captures the merchant’s side: what was offered, what obligations were accepted and what recourse exists.

The standard sets four levels of proof. At the lowest, an agent finds the terms and proceeding counts as consent. At the highest, a digital signature binds a named party to a specific document and hooks into arbitration, escrow and compliance systems. A hash of the terms travels with the payment record so both sides can prove what the document said at the moment of purchase.

Trust Stalls at Checkout

The consumer’s limits live in what the protocol calls a buyer policy. That policy declares spending caps, acceptable jurisdictions, accepted dispute bodies and the commitment level above which a person must review before the agent signs. Suggested triggers for that review include any charge above a set amount, any first-time merchant and any terms in a jurisdiction the shopper hasn’t dealt with before. When the policy escalates, the agent’s signing path halts until a human or a supervising agent approves.

That design tracks what shoppers already say they want. Consumers will let agents search across merchants, compare products and surface deals, but PYMNTS found trust drops as the agent nears payment, and they want approval rights and clear ways to cancel or reverse a purchase. According to Global Digital Shopping Index: The Agentic Commerce Deep Dive report, where PYMNTS Intelligence and Visa Acceptance Solutions surveyed 5,241 consumers, 1,185 merchants and 150 acquirers across the U.S., Brazil and the UAE, only 23% of U.S. consumers trust generative AI to handle payment transactions for them.

Retailers have started writing their own answers into the fine print. Target updated its terms of service in March so that once a customer authorizes an agent, any selection the agent makes is treated as the customer’s own. A policy like that by an airline would put the shopper on the hook for an airline the agent chose or a fare it accepted.

Bloomberg Intelligence expects about 20% of online commerce, or $500 billion, to run through agents acting on a single prompt by 2030, up from under 5% this year, Bloomberg Law reported. Another 6.2% will be fully autonomous, with agents deciding what to buy and when.