Identity Machines: Iron Book Trust Profile Extension for A2A Agents (v0.1)

[loganwolfe-ims]

We’re building Identity Machines — a zero-trust and security attestation platform for enterprise AI agents, centered around our Iron Book SDK/API, which acts as a security attestation and compliance layer for off-the-shelf and custom-built AI agents.

The idea is to help enterprises safely adopt AI agents by verifying and enforcing trust, compliance, and behavior — from build to runtime. We want to be the "Switzerland of Trust for AI Agents."

Long-term, the aim is to grow into a security-first marketplace for enterprise AI agents, underpinned by the Iron Book infra.

We are releasing Iron Book as a free GitHub Action shortly to drive developer adoption, incentivize AI security, and get feedback, as well as running free enterprise pilots to help in-house dev teams secure the agents they are already working on.

Why Developers and Enterprises Should Care

Developers: Boost enterprise adoption by making your agents trusted-by-default. Iron Book automatically generates a comprehensive UX/UI trust profile in a trusted agent registry for your agent builds (sample profile below).

Enterprises: Gain visibility and enforcement before and after deploy time. Only deploy and interact with agents meeting your trust, compliance, and security bar.

What We've Built Already

The current Iron Book SDK/API + developer tooling enables AI agent dev teams to:

• Secure their AI agents (in-house or third-party) using built-in vulnerability/CVE discovery, identity claims verification a-la OIDC, and runtime LLM behavioral weakness analysis with ~ 2,000 probes mapped to OWASP LLP Top 10.
• Attest to trustworthiness and compliance (SOC2, GDPR, PCI, etc.) with cryptographically signed Iron Book agent trust profile manifests, locked SBOM, and a trusted agent registry.
• Deploy agents safely into enterprise environments via self-managed (EKS, GKE) or hosted runtime (PaaS) with continuous trust checks. We call this the Iron Vow, and this is where A2A comes in.

Iron Vow Enhancement for A2A

Iron Vow is an extension of the Agent-to-Agent (A2A) protocol designed to enhance trust, security, compliance, and runtime integrity of AI agents.

• We extend the A2A agent.json / agent card spec with a trust_profile object.
• We provide an external signed attestation_url with deep security details.
• Iron Book API auto-generates and maintains these during agent onboarding.
• Enterprises use our lightweight verification layer at discovery, deploy, and runtime.

This enables third parties (such as enterprises) to programmatically assess the trustworthiness of an agent before or during interaction.

It also allows agents within multi-agent systems to verify each other's security, compliance, and trust parameters dynamically (alongside the capabilities and tooling) using trust-enhanced agent cards.

Simplified trust_profile Object Specification

"trust_profile": {
  "issuer": "Identity Machines Inc.",
  "trust_score": 92,
  "attestation_url": "https://trust.identitymachines.com/attestation/agent-abc123",
  "last_verification": "2025-04-28T10:42:00Z",
  "sbom_url": "https://trust.identitymachines.com/sbom/agent-abc123",
  "vulnerabilities": {
    ...
  },
   "llm-behavior": {
     ...
  },
  "compliance": {
    "pci_dss": "certified",
    "soc2_type2": "certified",
    "gdpr": "compliant",
    "hipaa": "non-compliant"
  },
  "runtime_monitoring": true,
  "integrity_protection": "cosign_signed",
  "revocation_status": "active"
}

Note: the cumulative trust_score calculation is based on customizable weights, e.g.:

Trust Score = (SAST_score * 0.3) + (Behavioral_score * 0.25) + (Compliance_score * 0.25) + (Integrity_score * 0.1) + (Revocation_score * 0.1)

The attestation package located at attestation_url is a signed JSON file containing OCI container digest (hash) for containerized agents, build timestamp, artifact metadata, verification results (vulnerability scan, runtime testing, compliance checks, etc.), and a rotated digital signature.

We're also working on integrating popular identity (e.g., AWS Cognito) and authentication (e.g., Auth0) frameworks to cover the zero-trust runtime access part of things for our trusted Agents, as well as to eliminate excessive agency / overprovisioning.

What We're Looking For

There's still a lot of work to be done, and we have a small and committed team. We'd love your help and participation to make this happen!

• Tell us what you think we're doing right
• Tell us what we could and should be doing better, or what's missing
• Request early access, more details, or a pilot
• Suggest collaboration, and ideas
• Ask questions

You can (and should!) also email us at devops@identitymachines.com with any suggestions, ideas, and requests.