Non-Human Identity.

A portable, cryptographic identity that every non-human principal carries. AI agents, service accounts, and the machines they command. Verifiable anywhere. Grounded in open standards.

What non-human identity is

Non-human identity is the category of identity that applies to anything that is not a human user. Traditional IAM was built for humans: employees signing in with SSO, customers holding accounts, contractors carrying badges. It handles the human case well. Non-human identity is the surface it was never designed for.

Every autonomous principal needs an identity it can carry across boundaries: AI agents, service accounts, workload identities, drones, robots, PLCs, controllers, sensors, and cyber-physical systems. That identity has to be portable across clouds and tenants, verifiable without calling home, and cryptographic rather than shared-secret. Non-human identity is the name for that primitive.

Why the term matters

Non-human identity is now a named category, with vendors staking out corners: SPIFFE and SPIRE for workload identity, Aserto for authorization at the identity edge, HashiCorp Vault for machine credentials, Astrix for the discovery and posture side. AI agents extend the frontier further. The term matters because it names the shape of the problem before any one vendor's approach.

The standards Veldt builds on

Veldt's identity primitives are grounded in open standards, not proprietary schemes. Interoperable with any system that also implements them.

• W3C DID (Decentralized Identifier). The self-sovereign identifier that no certificate authority controls. A DID is a principal's canonical name, resolvable to a public key, portable across any relying party.
• W3C VC (Verifiable Credentials). The signed claims about what a principal is authorized to do. A VC binds a scope, an issuer, and an expiry to a DID, and travels with the action.
• OIDC and OAuth 2.0. For federation into existing identity providers where they already exist. Non-human identity does not replace the human IdP, it extends past it.
• SPIFFE and SPIRE. Adjacent standards for workload identity, focused on the service-to-service case. Interoperable with the DID and VC layer for the broader principal set.

None of these are Veldt-invented. The value of an identity that any system can verify is that it is defined in the open.

What non-human identity looks like in practice

An AI agent starts a session. Its identity is a W3C DID, and its scope is a Verifiable Credential signed by the operator that stood it up. That credential travels with every action the agent takes. Every downstream system verifies the signature in a single round trip: no shared secret to leak, no API key to hardcode, no per-cloud identity silo.

The same shape applies to a service account calling a database, a drone accepting a mission update, a PLC pushing a state change. Each principal carries its own signed identity. The identity is the input to every access decision, not the output of a shared credential store.

Machines they command

When an AI agent orchestrates a drone, a PLC, a robot, or another agent, each downstream principal has its own signed identity. The delegation is a chain of signed credentials, not a chain of shared tokens. The drone's credential is signed by the agent, whose credential is signed by the operator, whose credential is signed by the human that started the workflow. Every hop is provable, every scope is bounded, every action attributable.

Why this matters for governance

Runtime governance can only enforce what it can attribute. If an action arrives with an ambient credential, the system knows only that some caller is authorized. If the action arrives with a signed non-human identity, the system knows which principal, with what scope, and under whose delegation. Identity is the input to enforcement.

This is why non-human identity sits under the Identity pillar and feeds into the Authority pillar. Without portable, signed identity, delegation attribution has nothing to trace, and per-action authorization has nothing to bind.

Non-Human Identity vs Traditional IAM

Non-human identity is not a replacement for IAM. It is the layer that extends IAM to the principals IAM was never designed for.

• Traditional IAM. Designed for humans. Session-based. Cloud-siloed. Credential rotation is human-triggered. Assumes a keyboard, a browser, and a slow-changing set of users.
• Non-Human Identity. Designed for autonomous principals. Per-action, not per-session. Portable across clouds and tenants. Credentials are cryptographic and short-TTL. Assumes a machine that acts on its own schedule.

The two coexist. Humans keep their IdP. Non-human principals get a DID and a VC. The two vocabularies meet at the point of delegation, where a human's authority becomes a signed credential that a non-human principal can carry.