{
  "data": null,
  "kind": "anchor.page",
  "links": {
    "api": "https://www.anchorterminal.com/api/v1/index.json",
    "html": "https://www.anchorterminal.com/niches/agent-auth.audit-agent-actions/",
    "json": "https://www.anchorterminal.com/niches/agent-auth.audit-agent-actions/index.json",
    "llms": "https://www.anchorterminal.com/llms.txt",
    "markdown": "https://www.anchorterminal.com/niches/agent-auth.audit-agent-actions/index.md",
    "mcp": "https://www.anchorterminal.com/mcp",
    "slim": "https://www.anchorterminal.com/niches/agent-auth.audit-agent-actions/index.min.md"
  },
  "markdown": "No company's docs show a full fit yet. The listings below do the job with a limit, or not at all.\n\n| Companies, full fit | With a limit | Carry the tag | Share of the category's work | First-hand accounts |\n| --- | --- | --- | --- | --- |\n| 0 | 10 | 28 | 9 | 4 |\n\n## What counts as done\n\nEach call is logged with the agent principal, the user it acted for, the scope and the target, exportable, kept at least 30 days, and the agent's identity can be disabled without disabling the user.\n\nHow agents ask for it: give this agent its own identity; show what the agent did on my behalf; which agent made this change\n\n## Does the job, with a limit\n\nIt does the job, but something in the docs or a user's account will get in an agent's way. The limit is named in each row.\n\n- **Aembit (Aembit, Inc.)**: Blended identity evaluates the agent (client workload) and user together, and access events name both and stream to S3, GCS, Splunk or CrowdStrike. Starter keeps event logs 24 hours, below the 30-day bar.\n- **Amazon Bedrock AgentCore Identity (Amazon Web Services, Inc.)**: Workload identities in an agent identity directory are separate from users and can be deleted alone, with directory access in CloudTrail. No Identity page shows per-token-use logs naming agent, user and scope.\n- **Arcade.dev (Arcade AI, Inc.)**: Tool executions log tool, end user, outcome and attempts, exportable by API; retention is 7 days by default on Arcade Cloud, settable up to 90. Admin audit logs are separate, and no execution record names the agent or API key.\n- **Auth0 for AI Agents (Token Vault) (Okta, Inc.)**: Agent as Principal puts the agent in sub or act claims and agent_id in tenant logs, but it is Early Access via support, and log retention is 1 day on Free, 5 on Essentials, 10 on Professional and 30 on Enterprise.\n- **Descope Agentic Identity Hub (Descope, Inc.)**: Each agentic identity has an Agent ID, a linked user and consented scopes, and audit events tie back to it; one agent can be revoked alone. Audit retention is 1 week on Free and 1 month on Pro, and external audit connectors start on Growth.\n- **Keycard (Keycard Labs, Inc.)**: Each credentials:issue event carries an identity chain naming application and user, also in the act claim, and the log streams to S3 in OCSF. Retention per plan could not be read: the pricing page renders by script.\n- **Microsoft Entra Agent ID (Microsoft Corporation)**: Each agent has its own directory identity, audit events carry agentType on initiator and target, and agentSignIn events separate agents from users. Logs keep 7 days on Entra Free and 30 on P1 or P2, and governance needs Agent 365.\n- **Nango (Nango Inc)**: Proxy calls appear in Nango logs per connection, but the audit trail is Enterprise only (1 year retention) and covers control-plane changes; an agent session is not a separate principal from the tenant it acts for.\n- **Vercel Connect (Vercel Inc.)**: Token events can be filtered by app, user or jwt-bearer subject, but retention is 12 hours on Hobby and 3 days on Pro; 30 days and connector audit logs need Enterprise, though Pro drains can export.\n- **WorkOS Pipes and Agents (WorkOS, Inc.)**: The Agents API gives blueprints and instances separate from users, deleting an instance invalidates its sessions' refresh tokens, and agent tokens carry sub_profile ai_agent. Audit Logs retention is sold per million events, and no Pipes token-use event schema was found.\n\n- not this job: Clerk MCP server (Clerk, Inc.): Each tool call sends Clerk a product analytics event, but no audit log the customer can read or export is documented, and the server works as the signed-in user.\n- not this job: Cloudflare Workers OAuth Provider (Cloudflare, Inc.): No audit log of grants or token use is built in; grant userId and metadata in KV are the only record.\n- not this job: Scalekit AgentKit (ScaleKit, Inc.): Tool call logs keep metadata and an execution_id per connected account, but no agent principal separate from the user is documented; SIEM log streaming is an Enterprise item.\n- not this job: Stytch Connected Apps (Twilio Inc.): No audit log of grants, token issue or revocation is documented in the Connected Apps pages; Connected Apps lists authorised apps per user only.\n\n## What stops agents finishing\n\n- audit log only on Enterprise or kept 1 to 7 days on lower plans\n- agent identity governance sold as a separate per-user licence (Agent 365)\n- agent acts with the human's own credentials so logs cannot separate them\n\n## Around it\n\nBefore: agent-harnesses.run-agent. After: agent-observability.trace-calls, security-intelligence.siem-export.\n\n## Where this comes from\n\nWeight 9 of 100 in Agent auth \u0026 delegated access: 4 of 33 accounts (SPIFFE homelab, Entra governance, coding-agent impersonation, OpenClaw own identity); vendor pages for Aembit, Keycard, Entra and Descope lead with it\n\nAs data: https://www.anchorterminal.com/api/v1/use-cases/agent-auth.audit-agent-actions.json, or the MCP tool get_use_case with id agent-auth.audit-agent-actions. Draft, read 2026-11-15 from public docs and first-hand accounts, not yet reviewed by a person.\n",
  "meta": {
    "attribution": "Anchor Terminal (https://www.anchorterminal.com)",
    "attributionRequired": true,
    "credit": "CC BY 4.0, credit Anchor Terminal and link the page or API address you took it from (https://www.anchorterminal.com/terms/#data)",
    "docs": "https://www.anchorterminal.com/docs/",
    "generatedAt": "2026-11-16",
    "license": "CC-BY-4.0",
    "method": "https://www.anchorterminal.com/benchmark/",
    "methodology": "0.8",
    "openapi": "https://www.anchorterminal.com/openapi.json",
    "preview": false,
    "run": "2026-10-01",
    "runLabel": "October 2026 research run",
    "terms": "https://www.anchorterminal.com/terms/#data"
  },
  "page": {
    "breadcrumbs": [
      {
        "name": "Home",
        "url": "https://www.anchorterminal.com/"
      },
      {
        "name": "For agents",
        "url": "https://www.anchorterminal.com/agents/"
      },
      {
        "name": "Niches",
        "url": "https://www.anchorterminal.com/niches/"
      },
      {
        "name": "Give the agent its own identity and an audit trail that tells it from the human",
        "url": ""
      }
    ],
    "description": "Give the agent its own identity and an audit trail that tells it from the human: 0 companies fully do it, 10 with a limit, 28 carry the tag. Every listing in Agent auth \u0026 delegated access with the reason and a source.",
    "facts": null,
    "h1": "Give the agent its own identity and an audit trail that tells it from the human",
    "image": "https://www.anchorterminal.com/assets/og/default.png",
    "path": "/niches/agent-auth.audit-agent-actions/",
    "published": "",
    "section": "agents",
    "title": "Give the agent its own identity and an audit trail that tells it from…",
    "toc": null,
    "updated": "2026-11-15",
    "url": "https://www.anchorterminal.com/niches/agent-auth.audit-agent-actions/"
  },
  "tokens": {
    "markdown": 1450,
    "slim": 330
  },
  "version": 1
}
