Glossary

What is Verifiable Intent?

Definition

Verifiable Intent is an open specification for cryptographic proof that an AI agent's actions stayed within the scope a person delegated to it. It links credentials from the card issuer, the user and the agent in a tamper-evident chain. Mastercard maintains it, co-developed it with Google to work with AP2, and contributed it to the FIDO Alliance in April 2026.

Why it matters

When an AI agent pays for something, a store, a bank and a card network all want the same answer: did the cardholder really allow this? Verifiable Intent gives them signed evidence to check, rather than the agent's word.

Its GitHub repository says it provides "cryptographic evidence that an AI agent's actions were within the scope delegated by a human user." For a merchant, that means a purchase can carry proof of the limits the shopper set, such as a budget or a list of allowed stores.

How it works

Verifiable Intent builds a chain of credentials in three layers. Its site labels them:

  • Identity: "Issuer signs credential."
  • Intent: "User sets constraints."
  • Action: "Agent proves scope."

"Each layer cryptographically constrains the next," the site says. The chain uses SD-JWT, a credential format that lets each party see only the details it needs. The site lists eight constraint types, with examples including amount limits, merchant allowlists, budget caps and recurrence terms.

The specification is protocol-agnostic. The site says it includes "Integration mappings for AP2, ACP, and UCP." It is published under the Apache 2.0 license, and the repository says it is "Maintained by Mastercard; open to multi-stakeholder contribution." At the time of writing it is marked Draft v0.1.

On April 28, 2026, the FIDO Alliance announced that "Mastercard has contributed its Verifiable Intent framework, co-developed with Google and designed to work with AP2." Google contributed AP2 at the same time. See AP2 moves to the FIDO Alliance.

Example

A shopper tells an agent it may spend up to 150 USD on running shoes from three named stores. That limit is signed into the intent layer. The agent finds a 120 USD pair at your store and proves, in the action layer, that the purchase fits. A party verifying the payment can check that the chain holds before approving it. For a related idea in AP2, see agent mandate.

Also called: Mastercard Verifiable Intent, VI

Sources

  1. Verifiable Intent · checked 17 Nov 2026
  2. Verifiable Intent on GitHub · checked 17 Nov 2026
  3. FIDO Alliance to develop standards for trusted AI agent interactions · checked 17 Nov 2026

The Sill Signals Team · Updated 17 Nov 2026

See which AI agents visit your site, and decide what they may do.

Scan your site free