Non-Human Identity (NHI)

Definition

Non-Human Identities (NHIs) are digital credentials assigned to software workloads — services, bots, automation scripts, and AI agents — rather than to human users. In the context of agentic AI, NHIs encompass every API key, service account, certificate, OAuth token, and JWT issued to an AI agent so it can authenticate to internal or external systems.

Aliases / Variants

  • Machine identity — broader term covering all software principals
  • Workload identity — often used specifically for cloud-native, certificate-based identity (e.g., SPIFFE SVIDs)
  • Service account — legacy term for IAM principals bound to a software service
  • NHI — preferred shorthand in recent vendor and analyst discourse

NHI taxonomy

Eleven distinct types, each with different rotation, ownership, and detection profiles (per What Are Non-Human Identities? (Oasis Security)):

TypeAuthentication mechanismRotation profile
Service AccountUsername/password or service-account secretDecoupled (secret separate from principal)
Service PrincipalClient secret or certificateDecoupled
System AccountOS-issued, typically high privilegeDecoupled
IAM Role (AWS-style)Temporary STS session credentialsDecoupled, short-lived
API KeyStatic keyDecoupled
Machine Identity (VM / container / serverless)Cloud-issued cert / SVIDDecoupled
OAuth TokenTime-limited bearerDecoupled
TLS / mTLS CertificateAsymmetric keyDecoupled
Storage Access KeyLong-lived, broad-permissionCoupled (see Identity-Credential Coupling)
SAS Token (Azure)Time-limited, granularCoupled
Personal Access Token (PAT)Developer-generatedOften coupled (the token IS the access)
Database UserApplication-levelDecoupled

The coupled rows are where the credential string IS the identity — rotation is identity rotation. See Identity-Credential Coupling for implications.

Why NHI Matters for Agentic AI

AI agent deployments dramatically amplify the NHI problem in three ways:

  1. Scale: Each agent (and often each agent task) may require one or more credentials. Large multi-agent systems can generate thousands of identities quickly.
  2. Ephemerality: Agents can be spun up and torn down dynamically; their credentials may not be systematically revoked.
  3. Autonomy: Unlike a human who can be asked to justify an action, an agent may exercise credentials in ways that are opaque without purpose-built tracing.

The result is an NHI estate that grows rapidly, is poorly inventoried, and is a high-value attack surface.

Scale evidence (triangulated)

SourceMetricYearType
CyberArk82:1 machine-to-human identity ratio2025Vendor (Vanson Bourne survey, n=2,600)
Rubrik Zero Labs45:1 average; up to 100:1 in some orgs2025Vendor
arXiv 2503.1825550K → 250K machine identities per enterprise (400% growth)2021–2025Academic preprint
SailPoint Horizons of Identity Security 2025–202669% of orgs have more machine than human identities; ~half deploy 10× more2025–2026Vendor (n=375 IAM decision-makers, methodology disclosed)
Verizon DBIR 2025Third-party breach involvement doubled YoY (15% → 30%), driven by ungoverned machine accounts2025Industry survey (annual, methodology disclosed)
GitGuardian / Verizon 2025441,780 exposed secrets in public repos; 39% web-app infra (66% JWTs)2025Industry-data quantitative
IBM 2025 Cost of a Data Breach10.22M US average; explicit recommendation for NHI controls2025Industry survey (Ponemon methodology)
ENISA Threat Landscape 2025Confirms identity- and credential-themed risk acceleration2025EU government

Different sources differ on the exact ratio (45:1 vs 82:1; SailPoint reports 69% of orgs have more machine than human, ~half 10×) but agree on the structural point: NHIs outnumber humans by an order of magnitude or more, and growth is exponential. The exact 82:1 figure is single-vendor and should be cited as CyberArk-specific; the directional claim is well-corroborated. See Source Triangulation Audit §Claim 1.

Why human-identity controls fail for NHIs

Eight structural differences make HR-driven IAM and credential-storage-focused PAM unsuitable (per What Are Non-Human Identities? (Oasis Security)):

PropertyHumanNHI
CentralizationIT-managed, single source of truthDecentralized; created by developers / IT / IaC
OwnershipTied to individualOften unowned; shared across teams / apps
ScaleLinear with headcount10–100× human count, growing exponentially
Rate of changeHR-driven (joiner/mover/leaver)Code-pace (per commit / per deploy)
ProvisioningIT-mediatedDeveloper-driven, often invisible to IT
Secret expirationFrequent password rotationOften never rotated; sometimes no expiration
Operational risk of rotationLowHigh (rotation breaks production workflows)
Authentication factor diversityThree-factor + MFA + SSOSingle-factor (the secret); no MFA equivalent

The implication is that legacy IAM/PAM cannot be retrofitted for NHIs at scale — purpose-built tooling is required.

Real-world incident anchors

IncidentVectorNHI lesson
Microsoft AI Storage BreachMisconfigured SAS token38TB internal data exposed (passwords + private keys); coupled identity-credential — see Identity-Credential Coupling
CircleCI BreachOAuth token compromiseMass-rotation across thousands of customer environments — rotation infrastructure must be tested
Mercedes-Benz BreachService accounts with excessive privilegesLong-lived, over-permissioned NHIs are persistent attacker access

Relationship to DSPM

The paper draws an explicit analogy: NHI vendors are to agent credentials what DSPM (Data Security Posture Management) vendors were to data sprawl. Just as DSPM helped enterprises discover and govern data that had proliferated across cloud stores without oversight, NHI tooling helps discover, classify, and govern the machine credentials that AI agents accumulate.

Framing from source

“Improper management of credentials related to NHIs is a people and process problem, not so much a technology problem.” — Insight Partners, 2025

Governance Capabilities

NHI governance solutions typically provide:

  • Discovery — inventory all machine identities across cloud, SaaS, and on-prem
  • Lifecycle management — rotation, expiry, and revocation of credentials
  • Scoped permissions — enforce least-privilege for each agent identity (extends into Identity Security Posture Management / ISPM)
  • OAuth / API scope governance — visibility and control over what OAuth tokens and API scopes agents have been granted

Credential Zero Problem

A foundational challenge: an agent must authenticate to a secrets vault or IdP to retrieve its working credentials. This bootstrapping step is the Credential Zero problem. Solutions include SPIRE certificates baked into the workload at deploy time, which avoid storing static secrets entirely.

Where It Appears