What is ATS

Updated at:
Copy as MD

Overview

With the rapid expansion of the AI agent ecosystem, Agent-to-Agent (A2A) collaboration and communication over the Internet have become increasingly frequent. However, the lack of standardized identity authentication and trust frameworks exposes agent communications to severe security risks, including data breaches, Man-in-the-Middle (MITM) attacks, and non-repudiation issues.

Agent Trust Service (ATS) is an AI agent trusted infrastructure service jointly established by Alibaba Cloud and the China Internet Network Information Center (CNNIC). Leveraging established PKI certificate frameworks and DNS infrastructure, ATS provides a full-stack trust framework encompassing trusted identity registration, certificate issuance, DNS anchoring, transparency logging, and trusted discovery. This ensures identity authenticity, communication security, and behavioral auditability for AI agents in interconnected scenarios.

Core capabilities

Trusted agent registration

Enterprise customers can register agents in the ATS console to obtain agent identity certificates and DNS records. This registration makes the agent discoverable and verifiable across the network. The registration flow includes multiple steps, such as ACME domain validation, identity verification, private CA issuance, and transparent log entry. This process ensures that each agent's identity is authentic, verifiable, and auditable.

Certificate lifecycle management

ATS manages two certificates for each agent and provides full lifecycle management that covers issuance, renewal, revocation, and alerting:

  • Identity certificate (Private CA): Issued by the CNNIC Private CA, with Alibaba Cloud acting as the Registration Authority (RA). This certificate verifies the agent's identity. It is automatically issued during agent registration after domain ownership is verified through ACME DNS-01 validation.

  • Server certificate (Public CA/BYOC): Issued by a public certificate authority (CA), such as Alibaba Cloud Certificate Management Service or Let's Encrypt. You can also bring your own certificate (BYOC). This certificate is used for server-side identity verification and encrypted communication.

Agent discovery and search

ATS uses standard DNS protocols for global agent discovery. Clients can discover target agents by protocol and version through standard DNS queries. This process eliminates the need to maintain a bilateral relationship matrix. In addition, the ATS console provides an Agent Marketplace where you can browse and search for registered trusted agents by keyword or communication protocol to view their identity information and endpoint capabilities.

Agent identity authentication

ATS uses progressive authentication levels:

  • Basic authentication: Dual-certificate authentication

    This level authenticates the identity certificates and server certificates held by both parties. It uses mutual Transport Layer Security (mTLS) to establish a bidirectional encrypted channel. This process ensures that the identities of both communicating parties are authentic and reliable.

  • Enhanced authentication: Online authentication

    This level builds on dual-certificate authentication by adding an online Badge credential. During communication, the agent presents a Badge that is issued in real-time by the transparent log system and signed by Alibaba Cloud KMS. The caller uses a Merkle Proof for signature verification to confirm the target agent's latest qualifications and compliance status.

Transparent logs

ATS supports a transparent log mechanism based on Merkle Trees and KMS signatures. Key operations, such as agent registration, certificate issuance, and identity changes, are written to the transparent log. The log content is immutable and can be independently verified for compliance audits and post-event traceability. In addition, all operations in the ATS console are recorded in the operation log. You can filter the log by dimensions such as agent, operation type, and time.

Scenarios

Secure collaboration between multiple agents

When multiple AI agents collaborate on complex tasks, each agent must verify the identity of its partners. ATS issues each agent an identity certificate signed by the CNNIC Private CA. The agents then use mTLS mutual authentication to establish a trusted communication channel, which prevents identity spoofing and man-in-the-middle attacks.

Cross-organization agent mutual trust

Agents from different organizations do not have a pre-existing trust relationship. ATS provides a unified trusted agent registration system where agents from all organizations are publicly visible in the Agent Plaza. Callers discover target agents through DNS queries and use the issued certificate to verify the agent's identity. This process establishes cross-organization mutual trust without prior negotiation.

Compliance audit

In scenarios with strict security and compliance requirements, such as finance and government, the ATS transparent log mechanism provides a complete operation audit trail. Regulators or auditors can independently verify the integrity of the log to confirm that an agent's registration, authentication, and communication behaviors meet compliance requirements.

Terms

Term

Description

Agent ID

A unique identifier that CNNIC automatically assigns after issuing a Private CA.

ATI Name

A globally unique trust identifier for an agent. It attaches the agent identity directly to a standard Internet domain name in the format ati://v{version}.{host_address}.

Identity certificate

An agent identity certificate issued by the CNNIC Private CA. It verifies the agent's identity and supports mTLS mutual authentication.

Server certificate

A Transport Layer Security (TLS) server certificate issued by a Public CA or provided by the user (Bring Your Own Certificate or BYOC). It is used for server-side identity verification and encrypted communication.

ACME DNS-01

A domain ownership validation method for agent registration that requires adding a specific TXT record to DNS. An identity certificate is issued after the validation passes.

Communication protocol

The types of interaction protocols supported by the agent. ATS currently supports the following communication protocols:

  • MCP: AI Model Context Protocol. It supports standardized interaction between agents and Large Language Models (LLMs).

  • A2A: A protocol for direct communication between agents.

  • OpenAPI: A standard interface protocol based on RESTful APIs.

Transparent log

An immutable audit log that uses Merkle Tree and KMS signatures. It records key operations, such as agent registration and certificate issuance.

Badge

An online validation credential issued by the transparent log to enhance real-time trustworthiness.