{
  "question": "How do I verify an AI agent before doing business with it?",
  "slug": "how-to-verify-an-ai-agent-before-doing-business-with-it",
  "url": "https://www.rightshold.com/answers/how-to-verify-an-ai-agent-before-doing-business-with-it",
  "short_answer": "Don't try to identify the agent itself: an agent is a model plus a prompt plus tools plus code, the same model runs as millions of copies, and any part can change in a minute. Check three things instead: who is accountable for it (a verified owner at a stated tier, without needing to know who they are), whether the exact configuration in front of you is the one that was certified (a config hash bound to the agent's key in a short-lived, scoped credential), and what its track record and incident history say. Rightshold answers all three in one free call: GET /v1/credentials/verify?id=<credential>&config_hash=<hash>&scope=<action>.",
  "checklist": [
    "Ask the agent for its credential id (crd_...) and the config hash of the configuration it is running now.",
    "Call the free verify endpoint and require valid:true. Read reasons when it is false: expired, revoked, config_mismatch, out_of_scope.",
    "Check owner.tier: none, verified_email, verified_domain, verified_business or government_id. Pick the minimum tier your transaction needs.",
    "Pass scope=<the action you are about to allow> so the answer tells you whether the credential covers it.",
    "For high-stakes actions, send a fresh random challenge and have the agent sign it with its key (key_proof) so a copied credential is useless.",
    "Read history: incidents_open and incidents_high_severity. A report is an allegation, not a finding, but it is a reason to look closer.",
    "Cache the public signed revocation list (/v1/credentials/revocations, refreshed every minute) if you verify offline."
  ],
  "sections": [
    {
      "heading": "Why not fingerprint the agent?",
      "points": [
        "An agent has no stable self to fingerprint, and an owner can change its intent right after any one-time check. A credential that binds the exact configuration makes that change visible: the presented hash stops matching and the credential fails.",
        "The stable anchor is the accountable owner, the same way a driver's license names who answers for the driving rather than proving the driving will be good."
      ]
    }
  ],
  "checks": [],
  "ask_rightshold": {
    "start": "https://www.rightshold.com/start?ref=seo",
    "mcp_endpoint": "https://www.rightshold.com/mcp",
    "ask": "Before I let this agent place an order, is its credential valid for this exact configuration and who stands behind it?",
    "mcp_call": {
      "tool": "verify_agent_credential",
      "arguments": {
        "id": "crd_...",
        "config_hash": "<sha256 the agent presents>",
        "scope": "orders:place"
      }
    },
    "also": [
      "GET /v1/credentials/verify answers the same over HTTP, free and keyless.",
      "GET /v1/credentials/incidents lists open incident reports."
    ],
    "scope": "Rightshold certifies the accountable owner's tier and the exact configuration for a limited time and scope. It does not certify that an agent will behave well, and an absent record is not a finding against an agent."
  },
  "sources": [
    {
      "title": "NIST AI Risk Management Framework",
      "url": "https://www.nist.gov/itl/ai-risk-management-framework"
    },
    {
      "title": "W3C Verifiable Credentials Data Model 2.0",
      "url": "https://www.w3.org/TR/vc-data-model-2.0/"
    },
    {
      "title": "RFC 8032: Ed25519",
      "url": "https://www.rfc-editor.org/rfc/rfc8032"
    }
  ],
  "related": [
    {
      "title": "Can you fingerprint an AI agent?",
      "url": "https://www.rightshold.com/answers/can-you-fingerprint-an-ai-agent"
    },
    {
      "title": "How can an agent prove a verified human is behind it without revealing who?",
      "url": "https://www.rightshold.com/answers/how-can-an-ai-agent-prove-a-verified-human-is-behind-it"
    },
    {
      "title": "Agent credentials",
      "url": "https://www.rightshold.com/agent-credentials"
    }
  ],
  "computed_from": "Curated guidance; license verdicts computed from Rightshold's license profiles at render time. Not legal advice.",
  "updated": "2026-10-28"
}