14 October 2026 GoDaddy, Infoblox, Google and Cloudflare Demonstrate Secure AI Agent Transactions Across Platforms. Read the press release →

Open reference architecture · Technical case study · Hands-on labs

Agents that have never met,
verifying each other.

Can AI agents from different organisations verify each other’s identity and authority with no prior arrangement between them? A completed flight purchase between independently deployed agents shows they can, on existing internet infrastructure and established trust mechanisms rather than centralised platforms.

scorecard · independent verifier · example run
  • PASS discovery
  • PASS publisher control
  • PASS registered identity
  • PASS instance identity
  • PASS mandate
  • PASS attribution
  • PASS revocation
  • REFUSED forged request
  • REFUSED tampered request

The problem

Neither agent can look the other up in its own directory

Many agent identity frameworks were written for agents inside one organisation, where a central system can issue, track and withdraw identities. Commerce between agents breaks that assumption. The buyer’s agent and the seller’s agent belong to different organisations, run on different platforms, and may never have met.

The experiment tests whether open standards and existing internet infrastructure can carry the identity, authentication and accountability work that a central system would otherwise do.

The experiment

Four agents, three clouds, four domains, no shared platform

A Madrid-to-Singapore flight purchase, completed by four independently deployed agents. Each party verifies the others through published records and cryptographic signatures. The person’s own AI assistant reaches the buyer agent over MCP, and simulated rogue agents test the seller’s refusals.

buyer.onbehalf.dev · AWS

Buyer agent

Works for the person. Finds the seller, checks it, collects signed prices, asks for the person’s approval and completes the purchase.

seller.skybound-travel.com · Cloudflare

Seller agent

Open to anyone who searches for a flight. Signs every price it quotes, and completes a checkout only under a mandate it can verify.

sealedby.dev · Google Cloud

Approval service

Where the person says yes. Shows what is being bought, applies the person’s own limits, signs the mandate the seller checks, and stops any attempt the person has not approved.

provedby.dev · Google Cloud

Verifier

The independent auditor. Repeats the other three’s checks from their public records, and publishes a signed verdict.

The steps

One purchase, eleven steps

  1. The person asks. The assistant turns the person’s request into a call to the buyer’s buy tool, with an access token issued after a passkey sign-in.
  2. The buyer finds the seller. Starting from the domain skybound-travel.com alone, the buyer finds the seller’s record in signed DNS.
  3. The buyer checks. Fourteen checks on the seller before sending the query.
  4. How to buy. The buyer reads the seller’s agent card, its commerce profile and its tool list.
  5. Search and signed quotes. The seller returns signed quotes; the buyer recomputes each fingerprint and opens a checkout.
  6. The person approves. With a passkey. The approval service signs two mandates.
  7. The buyer asks to buy. It closes both mandates and sends a signed completion request over UCP.
  8. The seller checks. Ten checks in code before touching any money.
  9. Payment and receipts. The payment is captured and three signed receipts come back.
  10. Records sealed. Each ledger is checkpointed in an auditable log.
  11. Independent check. The verifier compares the parties’ records and mandate chains, and confirms the payment.

See the eleven steps in detail · every check, one by one.

The standards

Each question answered by an open standard

Grouped by the question each one answers about identity, authentication or authority. Nothing was invented for this.

QuestionStandards
Who controls this name?DNS: SVCB, DNSSEC, DANE
Where is the agent, and what does it speak?DNS-AID
Who registered this agent?ANS
Which agents offer this capability?ARD
What does the agent say about itself?A2A agent card; MCP
How do I buy from it?UCP
Who sent this request, from which running copy?HTTP Message Signatures; SPIFFE and SPIRE
Did a person approve this purchase, within what limits?Passkeys; AP2; SD-JWT
Can anyone rewrite what happened?SCITT

What it shows

Meeting IMDA’s agent identity requirements

The case study maps the experiment to the agent identity best practices in IMDA’s Model AI Governance Framework for Agentic AI, where the agent on the other side belongs to a different organisation.

01

Identity: unique, accounted for, catalogued.

Each agent is named by a domain, with DNSSEC-signed records, and registered in ANS with an entry in a transparency log. Every identity check traces back to that name, and a lookalike domain inherits none of it.

02

Authority: scoped and bounded by the person.

Signed AP2 mandates cover one seller, one flight, one payee and one amount, for a bounded time, and work only with the buyer agent’s own key. The buyer cannot sign mandates itself.

03

Stopped at four levels.

The person can stop their agent or cancel one approval; the operator can withdraw the agent’s key or remove its DNS record. Every forged and tampered request was refused, and none led to a payment.

04

A record each side can hold the other to.

Each party keeps a hash-chained ledger, checkpointed in a public SCITT log. An independent verifier replays the checks from three domain names and publishes a signed scorecard.

What this does, and doesn’t, claim

  • It is a reference architecture, offered as a starting point, to show that interoperability is achievable among agents built across organisations and platforms. Agentic commerce is at an early stage.
  • A domain shows which organisation runs an agent, not which company stands behind the domain. A log proves that a record existed unaltered at a point in time; it does not vouch for the record’s content.
  • The ledgers are public here for illustration. In a commercial deployment the parties would keep them private and grant read access only to the verifiers they choose.
  • Payments ran in Stripe’s test mode; in the labs they are simulated. No real money moves.

Engage with us

Test it, and help the standards mature

  • Implement it yourself. The basic information, the labs, and the team on Discord when you get stuck.
  • Work with us. If you need a more guided approach, contact us and we can work with you.

Demonstrated at GovWare 2026 · Singapore · 13–15 Oct

Demonstrated by GoDaddy, Infoblox, Google and Cloudflare, in collaboration with Singapore’s IMDA.

InfobloxGoDaddy