OPEN PROTOCOL · ARIA 1.0
Every agent.
Verified.
Agent Registry for Identity & Authorization
The open protocol for AI agent identity. DNS-anchored from L1. Post-quantum native. Governed by a nonprofit.
ARIA here means Agent Registry for Identity & Authorization. It is unrelated to WAI-ARIA (accessibility).
AI agents are crossing organizational boundaries thousands of times per second — negotiating contracts, moving money, accessing sensitive data. There is no standard way to verify who they are, who authorized them, what they’re allowed to do, or how to stop them if something goes wrong.
Every platform invents its own answer. None of them talk to each other. None of them survive the agent crossing an organizational boundary.
We built the layer that was missing.
ARIA is an open protocol that gives every AI agent a verifiable, cryptographically-signed identity — anchored to the internet’s existing trust infrastructure and governed by a nonprofit. Anyone can verify its credentials.
AID — Agent Identity Document
Every agent registered through ARIA receives an AID: a W3C Verifiable Credential containing the agent’s identity, who authorized it, and the scopes it declares — cryptographically signed, portable, and independently verifiable. Including offline.
Four Trust Levels
Four levels are specified, from cryptographic self-service (L0) to a legal entity backed by someone with the power to bind it (L3). Domain binding through DNS begins at L1. Each level adds human and organizational verification on top of the same cryptographic foundation.
DNS Anchoring
Starting at L1, a credential names a domain its principal has bound with one DNS TXT record, published once for all its agents — the same infrastructure that has governed the internet for 40 years. L0 operates without DNS — the cryptographic entry point with zero friction.
Post-Quantum Cryptography
ML-DSA-65 + Ed25519 composite signatures (FIPS 204 + RFC 8032). Both must verify. Not a migration plan. The launch configuration.
ARIA AP — the identity, put to work
ARIA gives agents an identity. ARIA AP is how an agent uses it with a service: it checks who the receiver is, signs each request with its own key and gets a signed receipt for each operation. The receiver admits or refuses by its own rules.
Trust Ledger
Every credential lifecycle event — issuance, renewal, suspension, revocation, expiry, tombstone — permanently recorded in an append-only log. Not interaction logging. The audit trail that proves accountability.
Before a service acts, the agent’s request says exactly what it asks for — the operation, its content and the receiver — under the agent’s own signature. Accountability starts with a signed request.
ARIA does not replace any existing protocol. It provides the identity layer that none of them include.
A2A handles communication. SPIFFE handles infrastructure identity. MCP handles tool connectivity. OAuth handles tokens. Microsoft Entra, AWS AgentCore, and Google IAM handle cloud identity. None provide verifiable organizational identity with revocation across boundaries.
ARIA is the identity layer between trust domains — designed to complement, not compete.
When an agent arrives at your infrastructure, the protocol answers three questions in milliseconds.
Identity
Who is this agent? Who authorized it? Is the AID valid? Is it anchored to a domain the principal controls? Has it been revoked?
Request
What exactly is it asking for? The operation, its content and the receiver are covered by the agent’s signature, with a single-use challenge from the receiver: a copied credential cannot sign, and an altered request does not verify.
Admission
Does the level meet your rule? Is the scope one the agent declared and you admit? Is a required human approval present? ARIA AP checks them in order, then your own rule, and answers with a signed receipt.
Identity verification runs live at api.aria.bar. ARIA AP admission runs at each receiver that adopts it.
Maria runs a logistics company in Mexico City. She authorizes an AI agent to negotiate freight rates with three carriers. Each carrier’s system checks the agent’s ARIA credential: Is it genuine? Who answers for it? Has it been revoked? Can she revoke it?
At L0: the credential is genuine if it verifies under an issuer the carrier trusts, it names its principal (self-declared at L0), its status is published in a signed list, and Maria can revoke it; each carrier decides how fresh a status it accepts. Whether Maria’s company is a verified legal entity is what L2 adds.
Verification and revocation status are public at api.aria.bar; the organization behind an agent is verified from L2 up.
1,000 API requests per minute.
How many are agents?
No identity
Any process can claim to be anything. Without verified identity, every API call is blind trust.
Invisible scope creep
Agents exceed their mandate. Without boundaries, overreach is undetectable until damage is done.
No kill switch
Compromised agents keep operating. No standard mechanism to revoke credentials across systems.
Six protocol layers. One open standard.
Every layer maps to an existing standard. No new crypto. No proprietary formats.
Four trust levels. Progressive verification.
Every level post-quantum. Domain binding through DNS begins at L1. Behind every level: a human authorized this agent.
Cryptographic existence of the agent. Nothing more. Every declared name is marked self-declared.
E-mail challenge + key possession under ML-DSA-65 + Ed25519. Fully automatic.
Internal tooling, development, testing
A verified natural person answers for the agent, and the agent controls its domain. The organization remains self-declared.
L0 + person verification (ICAO 9303 document, ISO/IEC 30107-3 level 2 liveness, biometric match, human review) + domain via DoH (RFC 8484) + RDAP. Not automatic.
External agents, support bots, assistants
L1, plus: the entity exists and is active per its Authoritative Source, the person is linked to it, and neither is sanctioned.
Authoritative Source (Route A direct / Route B documentary, ≥20 % spot-check) + name normalization + person–entity link + OFAC/EU/UN screening ≤7 days.
B2B commerce, data access, procurement
L2, plus: a Binding Officer with the power to bind the entity backs the agent, keys live in certified hardware, signed accountability exists.
Primary commercial register + Binding Officer by public instrument + FIPS 140-3 L2+ / CC EAL4+ keys (MUST) + named human approval + signed accountability agreement.
Finance, healthcare, regulated contracts
ARIA gives agents a passport.
ARIA AP is how they use it.
The agent checks who it is talking to and signs each request with its own key. The receiver admits or refuses it by its own rules, runs each operation once and signs a receipt. Each side keeps its own evidence.
The agent checks the receiver’s signed identity against the one it expects at that address, then signs its request for that receiver, with a single-use challenge.
The receiver verifies the credential and its status, then audience, level, scope, delegation and human approval, then its own rule. A refusal is a signed receipt for that attempt only.
Once per operation, one signed receipt: completed, accepted or rejected. A retry gets the original; a lost answer never causes a second execution.
Trust Ledger — Everything is recorded
Append-only log of credential lifecycle events. The audit trail that makes agent identity accountable.
Behind every AI agent: a human.
At every trust level, a human principal authorizes the agent. L0: a human generated the keypair. L1: a verified person answers for the agent and it controls its domain. L2 and L3: the organization exists per its Authoritative Source and the person is linked to it, and at L3 a Binding Officer with the power to bind the entity backs it, with keys in certified hardware. The chain always leads back to a person.
Agents don’t exist in a vacuum. They exist because someone authorized them. ARIA makes that authorization cryptographically provable and revocable.
Try it right now.
curl https://api.aria.bar/v1/verify/did:aria:aria.bar:u-0dhNSrC9:commerce-search
Public endpoint. No authentication. It answers for the DID’s current credential.
import { createVerifier } from '@trustlayer-foundation/aria-sdk';
const verifier = createVerifier({
trust: [issuer], // the issuers you trust: none by default
fetch: { allowedOrigins: ['https://api.aria.bar'] },
});
const checked = verifier.verifyCredential(credential);
if (checked.ok) {
const status = await verifier.checkStatus(checked.credential);
console.log(checked.credential.did, verifier.standing(checked.credential, status).state);
}On npm · offline except the status list · TrustLayer Foundation · Apache 2.0. About the 1.0 packages →
Built on open standards. Filed with NIST.
Filed with NIST docket 2025-0035, March 9, 2026. NCCoE concept paper submitted April 2, 2026. Factual filings — not endorsements.
Owned by a nonprofit. Operated transparently.
Same model as Mozilla. Nobody owns the standard. Everybody can build on it.
TrustLayer Foundation A.C.
Owns the protocol, spec, and standard. Constituted March 20, 2026. Anti-capture clause. CC-BY-4.0 docs. Apache 2.0 code. Public RFC process.
Consejo: Aaron Grego (Presidente) · Adolfo Grego Micha (Secretario) · Ivan Moreno Mendoza (Vocal) · Carlos Grego Samra (Tesorero)
TUNO Labs SAPI de CV
Operates registry, API, platform, and SDK under ROLA license from TLF. Sibling entity, not subsidiary. 14 years DNS heritage via Punto 2012 (101K+ domains).
Technical Steering Committee
Technical authority on protocol and spec. Open membership through contribution. Apache/IETF model.
Working Groups
Community-driven, GitHub RFC process. Cryptography, trust levels, ARIA AP, commerce, compliance.
Advisory Board
AI infrastructure, identity standards, internet governance. 4-6 meetings/yr. Remote-first.
Open by design. Real by implementation.
ARIA is not a product. It’s infrastructure — and infrastructure belongs to everyone who builds on it.
Register your agent.
Today.
The agents are already here. The question is whether they have identity.