Definition
Verifiable Intent
Verifiable Intent is a cryptographic authorization protocol and credential framework designed to prove that a human user (the principal) has authorized a specific AI agent to take a specific action, such as a purchase or data transfer.
Updated
Definition
Verifiable Intent is a cryptographic authorization protocol and credential framework designed to prove that a human user (the principal) has authorized a specific AI agent to take a specific action, such as a purchase or data transfer.
As AI agents become increasingly autonomous in their ability to browse, shop, and transact, traditional identity verification—asking ‘who are you?’—becomes insufficient. Verifiable Intent addresses this gap by shifting the security model to ‘who authorized this agent to take this specific action?’ This protocol, co-developed by Mastercard and Google and launched in March 2026, establishes a standards-based framework where human users provide cryptographically signed, machine-checkable mandates to their agents.
The SD-JWT Delegation Chain
Verifiable Intent operates using Selective Disclosure JSON Web Tokens (SD-JWT), a technology that allows for the sharing of specific claims without revealing the entire underlying dataset. The protocol organizes these credentials into three layers bound in an SD-JWT delegation chain: an identity layer in which the issuer signs the credential, an intent layer in which the user sets the constraints, and an action layer in which the agent proves its actions stayed within scope. Because each layer cryptographically constrains the next, the agent’s behavior remains directly bound to the human principal’s original instructions, in a chain that is tamper-resistant end to end.
Machine-Verifiable Constraints
Unlike static policy documents or verbal instructions, Verifiable Intent credentials embed explicit limitations directly into the cryptographic record. The draft v0.1 specification registers eight machine-checkable constraint types, and verifiers must support all of them: merchant allowlists and product-selection limits (where and what the agent may buy); payee authorization (who gets paid); per-transaction amount ranges and cumulative budget caps (how much); merchant-managed and agent-managed recurring-payment terms; and a checkout-payment cross-reference binding each payment to the specific checkout it authorizes. Because these constraints are cryptographically bound, a merchant or service provider can verify compliance instantly without trusting the agent’s own assertions.
The Shopping Agent Analogy
Consider a user instructing a shopping agent to buy a specific coffee machine for no more than $150 from a list of three approved retailers. Instead of the agent simply presenting a payment method, it provides a Verifiable Intent credential. The retailer sees a cryptographically verified ‘handshake’ proving that the agent has a valid mandate from the user for this specific purchase at this price. This acts as a digital power of attorney, reducing the risk of rogue agent behavior and protecting the user’s payment credentials from unauthorized use.
KYA Interoperability
Verifiable Intent is a key component of the broader Know Your Agent (KYA) framework, an industry-wide effort to bring KYC-style compliance and trust to the agentic economy. Ant International, Mastercard, and Visa have begun collaboration on a KYA interoperability framework, aiming to make an agent verified on one network recognizable and trusted on another. The three organizations are working toward common principles across their respective protocols—Visa’s Trusted Agent Protocol, Mastercard Verifiable Intent, and Ant International’s Agentic Mobile Protocol—creating the foundation for a global, secure agentic marketplace.