Skip to main content
[cyphrs] A2A Identity

AI agents are the hidden multiplier of machine-to-machine trust.

Every autonomous agent needs a verifiable identity. Not an API key – a certificate. A2A identity is mTLS for the agentic era.

01 // The Hidden Multiplier

Every agent is an identity you didn't plan for

Your infrastructure was built for services with known lifetimes and predictable scaling. Agents change that equation.

Every new AI agent is a new identity to manage.

A single orchestration pipeline can spawn dozens of agents. Each one makes network calls, accesses APIs, and interacts with other agents.

Agents multiply faster than humans can track.

Agents spin up on demand, chain together autonomously, and call services you may not have anticipated. The identity surface area grows exponentially.

Without certificate-based identity, you're trusting agents implicitly.

Shared API keys can't distinguish one agent from another. There's no mutual verification, no revocation, no audit trail of which agent did what.

This is the same trust model failure – accelerated by AI.

The industry spent a decade learning that shared secrets don't scale for services. Agents are the same lesson, compressed into months.

02 // The Comparison

Why API keys fail at scale

API Keys
  • Shared secrets

    One key shared across multiple agents. Compromised once, compromised everywhere.

  • No mutual authentication

    The server checks the key. The client trusts blindly. No verification in either direction.

  • No instant revocation

    Rotating a key means finding every agent that uses it. In agentic systems, that's a moving target.

  • No audit of who used what

    When ten agents share a key, your logs show the key – not the agent. Attribution is impossible.

Certificate Identity
  • Unique per agent

    Every agent gets its own certificate. Identity is cryptographic, not a string in a config file.

  • Mutual verification

    Both sides prove identity before exchanging data. The agent verifies the service. The service verifies the agent.

  • Instant revocation

    Revoke one agent without affecting any other. CRL and OCSP propagate immediately.

  • Full audit trail

    Every issuance, every authentication, every revocation tied to a specific agent identity.

03 // Capabilities

What A2A Identity does

Certificate-based identity for autonomous agents. The same trust model that secures your services – extended to every agent in your infrastructure.

Unique identity per agent

Issue a certificate to every AI agent. Each identity is cryptographically unique – scoped to what the agent can access and how long it lives.

Mutual agent authentication

Agents prove identity to each other before exchanging data. Both sides of every interaction are verified – not just the caller.

Instant agent revocation

Revoke a compromised or misbehaving agent immediately. Every relying party knows within seconds – not at next key rotation.

Policy-driven trust boundaries

Define which agents can talk to which services, what actions they're permitted, and enforce those boundaries cryptographically.

Complete audit trail

Every agent credential issuance, authentication event, and revocation recorded. Who deployed the agent, when, and what it was authorised to do.

Framework-ready

Works with popular agentic frameworks including LangChain, CrewAI, AutoGen, and custom implementations. Standard X.509 – no proprietary SDKs.

04 // Architecture

How it works

Certificate-based identity for agents follows the same model as service identity – request, verify, issue, authenticate.

A2A Identity Flow
1
Request

Agent submits CSR with identity scope

2
Issue

Trust CA validates and signs certificate

3
Authenticate

Agent presents cert to peer agent or service

4
Verified

Mutual identity confirmed – data flows

05 // Use Cases

Where A2A Identity applies

Multi-agent orchestration

Agents proving identity to each other within orchestration pipelines. When Agent A delegates to Agent B, both sides verify before any data or instructions are shared.

Prevents agent impersonation and prompt injection through identity spoofing.

Agent-to-service authentication

Agents accessing internal APIs, databases, and infrastructure. Each agent authenticates with its own certificate – not a shared service account or API key.

Same mTLS handshake your services already use. No new protocol.

Regulated AI environments

Industries requiring audit trails and identity for every autonomous system. Financial services, healthcare, and government mandates demand knowing exactly which agent took which action.

Full issuance and revocation audit log for compliance reporting.

06 // Integration

Built on [cyphrs] Hub

A2A Identity isn't a separate product – it's a new identity class on the same infrastructure that manages your services and certificates.

mTLS

A2A Identity uses the same mutual TLS handshake that secures service-to-service communication. Agents are a new identity class – not a new protocol.

Trust CA

Agent certificates are issued by your own Trust CA. Same hierarchy, same policy engine, same revocation infrastructure. Agents get identities from a root you own.

Scout

Scout discovers agent endpoints across your infrastructure – identifying which agents exist, what certificates they hold, and which need identity provisioning.

Give every agent an identity

Early access is free. We're working with a small number of teams to shape the platform – including how agent identity fits into their infrastructure.