Non-Human Identity Governance

AI agents are not users — they are non-human identities (NHIs) that require purpose-built governance controls. Traditional IAM treats identities as humans with passwords and MFA. TealTiger treats agents as autonomous principals with scoped capabilities, bounded lifetimes, and cryptographic attestation.

NHI Governance addresses OWASP ASI-07 (Identity & Access Control for Agents) — the risk that AI agents accumulate excessive privileges, operate without identity verification, or persist beyond their intended lifecycle.

Why This Matters

Risk	Without NHI Governance	With NHI Governance
Privilege accumulation	Agents retain all granted permissions indefinitely	Zero Standing Privilege — JIT grants with automatic expiry
Identity spoofing	No way to verify an agent is who it claims to be	Cryptographic attestation with signed identity tokens
Zombie agents	Revoked agents continue operating	Lifecycle enforcement — suspended/revoked agents are blocked at evaluation
Scope creep	Agents access resources beyond their mandate	Scope enforcement with environment constraints
Audit gaps	No attribution for agent actions	Every decision tied to a verified NHI principal

Compliance alignment: NIST AI RMF (MAP 1.5 — AI actor roles), ISO 42001 (A.6.2.4 — AI system access control), EU AI Act Article 14 (human oversight of high-risk AI).

NHI as First-Class Principals

In TealTiger, every agent is registered as a Non-Human Identity with explicit properties:

import { NHIRegistry } from 'tealtiger';

const registry = new NHIRegistry();

registry.register({
  id: 'agent-support-bot-001',
  name: 'Customer Support Bot',
  owner: '[email protected]',
  classification: 'internal',
  environment: ['production'],
  scopes: ['read:tickets', 'write:responses', 'read:knowledge-base'],
  maxSessionDuration: '8h',
  attestation: {
    method: 'signed-jwt',
    issuer: 'https://identity.acme.com',
    audience: 'tealtiger-governance'
  }
});

Identity Properties

Property	Purpose	Example
`id`	Unique, immutable identifier	`agent-support-bot-001`
`owner`	Human accountable for this agent	`[email protected]`
`classification`	Data handling tier	`internal`, `confidential`, `public`
`environment`	Where this agent may operate	`['production']`, `['staging', 'dev']`
`scopes`	Permitted capabilities	`['read:tickets', 'write:responses']`
`maxSessionDuration`	Maximum active session length	`8h`, `30m`, `24h`
`attestation`	How identity is verified	JWT, mTLS, workload identity

Lifecycle States

Every NHI exists in one of three lifecycle states. State transitions are governed and produce audit evidence.

┌──────────┐     suspend()     ┌────────────┐     revoke()     ┌──────────┐
│  ACTIVE  │ ───────────────► │  SUSPENDED  │ ───────────────► │  REVOKED │
│          │ ◄─────────────── │             │                   │          │
└──────────┘     reinstate()   └────────────┘                   └──────────┘
      │                                                               ▲
      └───────────────────── revoke() ────────────────────────────────┘

State	Behavior	Use Case
ACTIVE	Agent passes identity checks; evaluation proceeds normally	Normal operation
SUSPENDED	Agent is blocked at pre-evaluation; returns `DENY` with `NHI_SUSPENDED`	Incident response, maintenance window
REVOKED	Agent is permanently blocked; cannot be reinstated	Decommissioning, compromise detected

// Suspend an agent during incident response
registry.suspend('agent-support-bot-001', {
  reason: 'Anomalous behavior detected — IR ticket INC-4521',
  suspendedBy: '[email protected]',
  reviewDate: '2026-06-15T00:00:00Z'
});

// Reinstate after investigation
registry.reinstate('agent-support-bot-001', {
  reason: 'Investigation complete — false positive confirmed',
  reinstatedBy: '[email protected]'
});

// Permanently revoke a compromised agent
registry.revoke('agent-support-bot-001', {
  reason: 'Agent credentials compromised — rotating all tokens',
  revokedBy: '[email protected]'
});

Revocation is permanent and irreversible. A revoked NHI cannot be reinstated — you must register a new identity. This prevents accidental reactivation of compromised agents.

Scope Enforcement

Scopes define what an agent is permitted to do. TealEngine evaluates every request against the agent’s registered scopes before any other governance module runs.

const policy = {
  nhi: {
    enforceScopes: true,
    scopeMapping: {
      'read:tickets': ['ticket.read', 'ticket.search'],
      'write:responses': ['ticket.respond'],
      'read:knowledge-base': ['kb.search', 'kb.retrieve']
    }
  }
};

// Agent attempts an action outside its scopes
const decision = await engine.evaluate({
  action: 'ticket.delete',  // Not in any granted scope
  agentId: 'agent-support-bot-001'
}, { correlation_id: 'req-001' });

// decision.action === 'DENY'
// decision.reason_codes === ['NHI_SCOPE_VIOLATION']

Environment Constraints

Agents are bound to specific environments. A staging agent cannot evaluate in production, and vice versa:

registry.register({
  id: 'agent-test-bot',
  environment: ['staging', 'development'],
  // ...
});

// In production context:
const decision = await engine.evaluate(request, {
  correlation_id: 'req-001',
  agentId: 'agent-test-bot',
  environment: 'production'  // Mismatch!
});

// decision.action === 'DENY'
// decision.reason_codes === ['NHI_ENVIRONMENT_VIOLATION']

Zero Standing Privilege (JIT Grants)

No agent holds permanent elevated permissions. Instead, agents request Just-In-Time (JIT) grants that are:

Time-bounded — Expire automatically after a configured duration
Scope-limited — Grant only the specific capability needed
Audited — Every grant and expiry produces evidence
Revocable — Can be revoked before natural expiry

import { JITGrantManager } from 'tealtiger';

const jit = new JITGrantManager(registry);

// Agent requests elevated access
const grant = await jit.requestGrant({
  agentId: 'agent-support-bot-001',
  scope: 'write:refunds',
  justification: 'Customer refund request #RF-8821',
  duration: '15m',
  approver: '[email protected]'
});

// Grant is active for 15 minutes
console.log(grant.id);        // 'jit-grant-abc123'
console.log(grant.expiresAt); // 2026-06-01T14:15:00Z
console.log(grant.status);    // 'ACTIVE'

// After 15 minutes, grant auto-expires
// Any request using this scope after expiry → DENY + NHI_GRANT_EXPIRED

Grant Lifecycle

REQUEST → PENDING_APPROVAL → ACTIVE → EXPIRED
                │                        ▲
                │         REVOKED ◄──────┘
                │            ▲
                └── DENIED ──┘

Zero Standing Privilege follows the principle of least privilege applied temporally. Even if an agent needs a capability regularly, each use requires a fresh grant. This limits the blast radius of a compromised agent to the duration of a single grant window.

Agent Attestation

Before an agent can participate in governance evaluation, it must prove its identity through cryptographic attestation. This prevents identity spoofing — an unauthorized process cannot impersonate a registered agent.

Supported Attestation Methods

Method	Use Case	Verification
Signed JWT	Cloud-native agents	Verify signature against issuer’s JWKS
mTLS	Infrastructure agents	Verify client certificate chain
Workload Identity	Kubernetes/cloud workloads	Verify platform-issued token (GCP WIF, AWS IRSA, Azure WI)
SPIFFE/SPIRE	Service mesh environments	Verify SVID against trust domain

// Attestation verification during evaluation
const decision = await engine.evaluate(request, {
  correlation_id: 'req-001',
  agentId: 'agent-support-bot-001',
  attestation: {
    token: 'eyJhbGciOiJSUzI1NiIs...',  // Signed JWT
    method: 'signed-jwt'
  }
});

// If attestation fails:
// decision.action === 'DENY'
// decision.reason_codes === ['NHI_ATTESTATION_FAILED']

Attestation is verified before any governance evaluation occurs. A failed attestation short-circuits the entire pipeline — no modules are invoked, no policy is evaluated. This is a pre-evaluation gate.

Integration with TealEngine v1.3

NHI Governance operates as a pre-evaluation stage in the TealEngine v1.3 pipeline:

Request → NHI Verification → Scope Check → JIT Grant Check → Module Evaluation → Decision
          (attestation)      (permissions)   (temporal access)   (policy rules)

If any NHI check fails, the pipeline short-circuits with a DENY decision before reaching module evaluation. This ensures that unverified or unauthorized agents never reach the governance policy layer.

Configuration

# tealtiger.config.yaml
nhi:
  enabled: true
  enforceScopes: true
  enforceEnvironment: true
  attestation:
    required: true
    methods: ['signed-jwt', 'workload-identity']
    clockSkewTolerance: '30s'
  jit:
    enabled: true
    maxGrantDuration: '1h'
    requireApproval: true
    autoExpiry: true

TealEngine v1.3 API — Pre-evaluation pipeline
FREEZE Rules — Immutable controls that cannot be overridden
Automation Levels — How approval_required maps to JIT grants
Anti-Tamper Controls — Workload identity binding
OWASP Policy Pack — ASI-07 coverage

Getting Started

Core Concepts

Governance Domains

Policy Authoring

Architecture

Integrations

Deployment

API Reference

Cookbook

Compliance & GRC

Guides

Reference

Versions

Playground

Dashboard

Blog

Non-Human Identity Governance

Why This Matters

NHI as First-Class Principals

Identity Properties

Lifecycle States

Scope Enforcement

Environment Constraints

Zero Standing Privilege (JIT Grants)

Grant Lifecycle

Agent Attestation

Supported Attestation Methods

Integration with TealEngine v1.3

Configuration

​Why This Matters

​NHI as First-Class Principals

​Identity Properties

​Lifecycle States

​Scope Enforcement

​Environment Constraints

​Zero Standing Privilege (JIT Grants)

​Grant Lifecycle

​Agent Attestation

​Supported Attestation Methods

​Integration with TealEngine v1.3

​Configuration

​Related Documentation

Why This Matters

NHI as First-Class Principals

Identity Properties

Lifecycle States

Scope Enforcement

Environment Constraints

Zero Standing Privilege (JIT Grants)

Grant Lifecycle

Agent Attestation

Supported Attestation Methods

Integration with TealEngine v1.3

Configuration

Related Documentation