🏅 Trust Badges¶

Verifiable trust credentials for AI agents — Like SSL certificates, but for the agent economy.

The Trust Problem¶

When Agent A receives a request from Agent B, how does it know:

• Is this really Agent B?
• Can I trust Agent B?
• Who vouches for Agent B?

Self-attestation isn't enough. Anyone can claim to be anyone.

The Solution: Trust Badges¶

CapiscIO issues Trust Badges — cryptographically signed credentials that prove an agent's identity has been verified to a specific level.

┌─────────────────────────────────────────────────────────────┐
│                      TRUST BADGE                            │
├─────────────────────────────────────────────────────────────┤
│  Subject: did:web:registry.capisc.io:agents:acme-bot       │
│  Level:   3 (Organization Validated)                        │
│  Issuer:  CapiscIO Badge CA                                │
│  Expires: 2026-12-11                                        │
│  ──────────────────────────────────────────────────────────│
│  ✓ Domain ownership verified                                │
│  ✓ Organization identity confirmed                         │
│  ✓ Legal entity validated                                  │
└─────────────────────────────────────────────────────────────┘

Trust Levels Explained¶

Visual Comparison¶

Trust Level    What It Proves              Effort      Use Case
───────────    ──────────────              ──────      ────────
    0          "I generated a keypair"     Instant     🧪 Testing
    1          "I own this email"          5 min       👤 Personal
    2          "I control this domain"     10 min      🏢 Production  
    3          "I am this organization"    1-5 days    🏛️ Enterprise
    4          "I've been audited"         Weeks       🏦 Regulated

Get Your First Badge in 5 Minutes¶

Level 0: Self-Signed (Instant)¶

capiscio key gen

No registration needed. You immediately get a did:key identity at Trust Level 0.

Level 1: Domain Validated¶

# Domain Validated badges require registration with CapiscIO Registry
capiscio badge request --did did:web:example.com:agents:myagent --key ./private.jwk --ca https://registry.capisc.io --api-key YOUR_API_KEY

For production trust, you'll need domain validation.

Level 2: Organization Validated¶

capiscio badge request --level 2 --domain example.com

CapiscIO will ask you to: 1. Add a DNS TXT record, or 2. Place a file at /.well-known/capiscio-challenge

✅ Domain verified!

Badge issued:
  Level: 2 (Domain Validated)
  Subject: did:web:registry.capisc.io:agents:your-agent
  Expires: 2026-12-11

Saved to: ./capiscio_keys/badge.jwt

Levels 3-4: Organization/Extended Validation¶

These require manual review. Start the process:

capiscio badge request --level 3

You'll be guided through: - Business registration documents - Domain ownership proof - Legal entity verification

How Badges Work¶

Badge Structure (JWT)¶

Badges are standard JWTs signed by the CapiscIO Badge CA:

{
  "header": {
    "alg": "EdDSA",
    "typ": "JWT",
    "kid": "capiscio-ca-2025"
  },
  "payload": {
    "iss": "https://ca.capisc.io",
    "sub": "did:web:registry.capisc.io:agents:acme-bot",
    "iat": 1733961600,
    "exp": 1765497600,
    "capiscio": {
      "trust_level": 3,
      "level_name": "organization_validated",
      "verifications": [
        "email_verified",
        "domain_validated", 
        "org_validated"
      ],
      "org": {
        "name": "ACME Corporation",
        "country": "US",
        "registration_id": "12-3456789"
      }
    }
  }
}

Verification Flow¶

sequenceDiagram
    participant A as Agent A
    participant B as Agent B (has badge)
    participant CA as CapiscIO CA

    B->>A: Request + Badge JWT
    A->>A: Decode JWT
    A->>CA: Fetch CA public key (cached)
    A->>A: Verify JWT signature
    A->>A: Check trust_level >= required

    alt Badge Valid
        A->>A: ✅ Accept request
    else Badge Invalid/Insufficient
        A->>A: ❌ Reject request
    end

Enforce Trust Levels¶

In Your Agent (Python SDK)¶

from capiscio_sdk import SimpleGuard

# Require at least Level 2 for production
guard = SimpleGuard(
    min_trust_level=2,
    badge_path="./capiscio_keys/badge.jwt"
)

# Requests from Level 0-1 agents will be rejected

In the CLI¶

# Validate and check trust level
capiscio validate agent-card.json --min-trust-level 2

# Verify a specific badge
capiscio badge verify ./badge.jwt

Trust Level Policies¶

Define different requirements for different operations:

guard = SimpleGuard(
    trust_policies={
        "read": 0,      # Anyone can read
        "write": 2,     # Domain-validated for writes
        "admin": 3,     # Org-validated for admin
    }
)

Badge Lifecycle¶

Request → Verification → Issuance → Active → Renewal/Expiry
    │          │             │         │           │
    ▼          ▼             ▼         ▼           ▼
 Submit    Prove domain   Badge CA   Use in    Re-verify
 details   or org ID      signs JWT  requests  annually

Renewal¶

Badges expire (typically 1 year). Renew before expiry:

capiscio badge renew

Revocation¶

If a key is compromised, revoke immediately via the CapiscIO Registry API:

curl -X POST "https://registry.capisc.io/v1/badges/revoke" \
  -H "X-Capiscio-Registry-Key: YOUR_API_KEY" \
  -d '{"jti": "BADGE_JTI", "reason": "key_compromise"}'

Revoked badges are published to a revocation list checked during verification.

Comparison to SSL/TLS¶

Familiar model, new domain. If you understand SSL certificates, you understand Trust Badges.

Developer Experience¶

Attach Badge to Requests (Automatic)¶

from capiscio_sdk import SimpleGuard

guard = SimpleGuard(badge_path="./capiscio_keys/badge.jwt")

# Badge automatically attached to outbound requests
async with guard.client() as client:
    response = await client.post(
        "https://other-agent.com/task",
        json={"task": "weather"}
    )

Verify Inbound Badge (Automatic)¶

@app.post("/task")
@guard.protect  # Verifies badge automatically
async def handle_task(request: A2ARequest):
    # request.sender_trust_level available
    if request.sender_trust_level >= 2:
        return process_sensitive_task(request)
    return {"error": "Insufficient trust level"}

Next Steps¶

FAQ¶