Partner · Blockchain Life 2026 — Dubai, December 1–2 · 15,000+ attendees from 130+ countriesGet tickets →
LIVE
BTC—ETH—SOL—BNB—XRP—ADA—AVAX—DOGE—LINK—DOT—MATIC—ATOM—LTC—TRX—TON—BTC—ETH—SOL—BNB—XRP—ADA—AVAX—DOGE—LINK—DOT—MATIC—ATOM—LTC—TRX—TON—
—▲0.0%
Web3

Agent Identity Verification: How AI Agents Authenticate Purchases

4 min readGuide
Agent Identity Verification: How AI Agents Authenticate Purchases

Agent identity verification is the check a seller runs before letting a piece of software buy something. It answers three questions at once: which agent is this, which person authorised it, and what was it allowed to spend? The answer arrives as a signed credential the merchant can verify, rather than a promise typed into a form.

Why isn't a card number enough?

A card number says an account exists. It says nothing about who is holding it, and to a merchant's fraud system an agent buying headphones looks exactly like a script replaying a stolen token. Think of it as a permission slip rather than a key: the slip names the agent, names the person who signed it, and states the limits, so the shop can read it at the door.

How is the proof actually built?

Four approaches are in use, and they overlap rather than compete:

  • Tokens tied to a registered agent. Mastercard Agent Pay issues a credential binding the cardholder, the enrolled agent and the spending scope together.
  • Attestation headers. Visa's Trusted Agent Protocol has the agent attach a verifiable statement of who it is and what the user agreed to.
  • Signed mandates. Google's Agent Payments Protocol, open-sourced with more than sixty partners, has you sign an intent mandate covering what the agent may buy and a cart mandate fixing the exact items and price.
  • Decentralized identifiers. A crypto-native agent gets a DID - a public identifier anyone can look up to find its key. It signs with that key, and the counterparty checks it without asking a central registry. This is the model used alongside stablecoin payments on Base and Solana.

What does this stop, and what doesn't it?

It stops impersonation and replay: a credential is tied to one agent and one authorisation window, so an old one picked up later is worthless. It stops an agent spending outside the limits you signed. It does not stop an authorised agent from being confidently wrong, buying the right thing twice or the wrong thing once, and a stablecoin payment does not reverse.

The practical habit, if you let an agent loose: read the mandate before you sign it, because the amount, the time window and the list of what it may buy are the only real brakes. Keep the agent's money in its own wallet or a card with just that balance on it. Next: how agents pay for APIs and agentic payments explained.

← All guides