Skip to main content
Oathera logo Oathera
Use casesFive stories: no identity, a new identity, impersonation, cloning, and a call outside the boundary. PlatformOathera Agent Enrollment, the Oathera Identity Service and Oathera Agent Enforcement. FeaturesWhat Oathera does for you, grouped by purpose. IntegrationsYour identity provider, your MCP servers, your cloud. SecurityHow Oathera fails closed, and what it doesn't do.
Developer sectionComponents, the identity token, standards and settings. Data sheetsThe buyer and developer data sheets, as PDFs. LearnGuides on giving AI agents an identity of their own. GlossaryPrincipal, sponsor, anchor key, key digest, identity token. SimulatorWatch a stolen token, a replayed proof and a tampered call get refused.
About usThe team and the mission behind Oathera. BlogPosts from the Oathera team. CareersOpen roles at Oathera. ContactTalk to the founders, or become a design partner. LinkedIn ↗Oathera on LinkedIn.
Privacy PolicyHow we handle data. Terms of ServiceTerms for using Oathera. Data Processing AgreementOur DPA for customers. Sub-processorsThird parties we rely on.
Become a design partner

Glossary

The Know Your Agent vocabulary, in plain language.

Oathera draws lines that others blur. This glossary gathers every term in one place. Each term has its own anchor, so a definition anywhere on the site can link straight to it.

Become a design partner
Close-up of a dictionary page showing the entry for the word glossary

On this page

  1. All terms

All terms

Every Oathera term, defined in one place and listed by topic. Each term has its own anchor, so a definition anywhere on the site can link straight to it.

Know Your Agent (KYA)

The category: establishing which AI agent is acting, who stands behind it, and what it may do, by analogy with Know Your Customer. Know Your Agent is the category, not a product name. Oathera is the company and the product.

Non-Human Identity (NHI)

A non-human identity (NHI) is the identity of a machine actor — an AI agent, service, or workload — rather than a person. For AI agents, Oathera makes NHI a cryptographic, machine-bound signature: an anchor key created on the agent's own host, bound to the machine it runs on, so the agent cannot be cloned, impersonated, or run outside its authorized boundary. This is the foundation for agentic identity control and autonomous agent governance.

Oathera says machine-bound non-human identity, not a shared secret or bearer token: holding a credential is never enough.

Machine-Bound NHI vs. Traditional IAM

Traditional IAM issues shared secrets and bearer tokens that work for anyone who holds them, from any machine, so copying the credential copies the identity. Machine-bound NHI ties an AI agent's identity to a key created on its host that never leaves the machine; the host is recorded at enrollment and compared at every token, so a cloned agent is caught and a stolen token proves nothing without the key held only in memory. In short: IAM was built to authenticate people and their long-lived accounts, while machine-bound NHI is built to govern autonomous agents on every call.

Agentic Identity Control

Agentic identity control is the practice of proving which autonomous AI agent is acting, who approved it, and what it may do — on every operation, not once at login. Oathera enforces it with a machine-bound non-human identity, proof of possession on every call, and a signed boundary decision at Oathera Agent Enforcement, so an agent's authority is checked continuously rather than assumed from a token.

Agentic Governance Space

The agentic governance space is the emerging category of controls for securing and governing autonomous AI agents and their non-human identities: enrollment and approval, machine-bound identity, runtime checks, boundary enforcement, and accountability back to a named person. Oathera operates in this space as the identity and governance layer for AI agents, alongside — not in place of — your identity provider, egress controls, and AI governance.

Sponsor

The party accountable for an agent: a person or an organization. The sponsor may be someone other than the principal who approved the agent. The enrollment attribute block carries a sponsor field for it.

Oathera says sponsor, not the generic “owner.” A sponsor is accountable for the agent; it is a distinct idea from the principal who signs in to approve it.

Principal

The person who answers for an agent: the person who signed in and approved its enrollment, at the company's identity provider or in the Oathera console. Every principal is confirmed at enrollment. A principal is not the same as a sponsor: the principal performs the approval, while the sponsor is who the agent answers to.

Anchor key

The long-lived Ed25519 key created and kept on the agent's host. One agent has exactly one anchor key, and the agent's identity is the key digest of this key.

Oathera says anchor key rather than “the agent's private key”, which could mean the session key too.

Key digest

The hash of a public key. An agent's key digest is the hash of its anchor key, and each identity token names its session key the same way. With the host digest and the operation proof, it is part of what the agent leaves on every call.

Oathera says key digest, the name that pairs with host digest. On the wire the claim stays jkt, the standard name.

Identity token

The short-lived pass the Oathera Identity Service issues to an agent, naming the session key that may use it. It is issued only after the host is compared with the enrollment record, and lasts five minutes at most. Without the session key it is useless.

Oathera says identity token, or “the pass”, not “bearer token”: holding the token is not enough.

Session key

A short-lived key, replaced at every token issuance and never written to disk, that signs the agent's operation proofs.

Operation proof / Proof of possession

The operation proof is a single-use signature by the session key over the exact operation, its arguments and the identity token, with a nonce that is used once. Proof of possession is the property it gives: a receiver accepts the token only when the caller proves it holds the matching key, so a call can't be replayed or altered. The operation proof, the key digest and the host digest together make your agent's fingerprint: the mark it leaves on every call. A copy of the agent on another machine has a different host, so it leaves a different fingerprint.

Host binding

The policy that compares the host digest at each token issuance with the one recorded at enrollment: off, warn, or enforce. Enforce refuses a mismatch. Warn, the default for testing and design partners, issues the token and records a high-severity alert.

Oathera says host throughout, not a forced synonym; incidental technical phrases such as “virtual machine” are left as they are.

Host digest

A digest over the host attributes that should not change while an agent stays on one machine. It is computed on the agent's machine at enrollment and again at each token request. It is a digest of the host, not a claim that the host has been independently attested.

Oathera says host digest, not “host fingerprint”.

Enrollment

Registering an agent's anchor key with the Oathera Identity Service, approved by a principal. The anchor key is created on the agent's host, and a host digest is recorded. The Oathera Identity Service keeps everything it learns as the enrollment record, which holds your agent's DNA: what the agent stated about itself, the principal who approved it, its NHI attributes, and what was found on its host. The full record stays with the Oathera Identity Service; each identity token carries a summary of it as claims.

Enrollment approval

A principal signing in through the identity provider to approve an agent's enrollment. The approval is part of the agent's enrollment record.

Operational boundary

What an agent may do: which operations, on which audiences, at what clearance. The boundary carries the requested clearance and a list of permitted services.

Boundary decision

The Oathera Identity Service's signed, single-use allow or deny for one operation, requested by Oathera Agent Enforcement. Oathera Agent Enforcement refuses the call if no valid decision arrives.

Control plane

The Oathera-hosted services that enroll agents, issue tokens and hold policy. The control plane is distinct from the enforcement plane that acts beside your agent.

Enforcement plane

Oathera Agent Enrollment and Oathera Agent Enforcement, which sign and verify each request inside your environment.

Oathera pairs the control plane with the enforcement plane; it does not use the term some vendors use for the layer that carries traffic.

Oathera Witness

One of Oathera's three components: a small service on every machine where agents run, your servers and laptops, installed by IT. It watches the agents there from the operating system and vouches for what it saw in a signed report that Oathera Agent Enforcement checks before policy uses it.

Oathera MCP Plugin

One of Oathera's three components, with two capabilities on your MCP server: Oathera Agent Enrollment and Oathera Agent Enforcement.

Oathera Agent Enrollment

One of the Oathera MCP Plugin's two capabilities, on your MCP server: it gives each agent its identity, recording its own key, the host it runs on and the person who answers for it. The agent's key, and the proof it signs for each call, stay on the agent's own host.

A capability of the Oathera MCP Plugin. Oathera says Oathera Agent Enrollment, one name everywhere, not a generic “helper.”

Oathera Agent Enforcement

The check in the customer's MCP server, in front of its tools. It verifies the identity token and the operation proof, requests the boundary decision, and lets a request through only when it is allowed.

A capability of the Oathera MCP Plugin. Oathera says Oathera Agent Enforcement, one name everywhere, not a generic “gateway” label.

Oathera Identity Service (OIS)

One of Oathera's three components, in Oathera's cloud: it holds the register of approved agents, issues identity tokens and records every decision.

Credential service

The Oathera service that issues verifiable credentials recording who approved what, when the Oathera Identity Service is set to issue them.

Technical name: VCI.

Runtime check-in

Any point after enrollment where the enforcement plane consults the control plane: each token request and each boundary decision. When renewal stops, access ends within one token lifetime.

Principal deprovisioning

Stopping every agent a principal approved, without touching the person's own sign-in. The Oathera Identity Service exposes an endpoint for it. Revoking a single agent stops that agent alone.

Content trust

The label on content (trusted, agent or untrusted) that stops tool output and model-supplied arguments from widening an agent's authority. It is not a prompt-injection scanner.

Federation

Issuing tokens that another identity system accepts on an agent's behalf. For Microsoft Entra, the Oathera Identity Service issues a token that Entra accepts.

NHI attributes

The agent's answers to the OWASP non-human identity questionnaire, from which its risk tier is judged. Self-reported, not independently proofed.

Risk tier

A rating from 1 to 4 for how dangerous an agent's reach is, where 1 is the most dangerous and 4 the least. Risk tier is the only Oathera term that uses that word.

Resource server

The customer's tool or system behind Oathera Agent Enforcement. Oathera ships a sample one for demos.

Protected core

The signed native library inside both MCPs that holds the enforcement logic. It is checked before it loads and while it runs, and the MCP stops if it has been changed.

IdP

Your own identity provider, where a principal signs in over OIDC. Oathera relies on your IdP for human identity. The IdP in Oathera's hosted demo is a demo IdP, for demonstration only.

Capability

A named action an agent may perform, fixed into the identity token at issuance.

Privileged capability

A capability that grants a write or administrative action, and so shortens the identity token's lifetime.

Audience

The one resource an identity token is minted for, so a token meant for one resource server is not accepted by another. The gateway assertion names one resource audience.

Anchor rotation

Moving an agent to a new anchor key without enrolling again.

Gateway assertion

Oathera Agent Enforcement's signed, single-use statement to the resource server that one request passed its checks, sent in place of the agent's token.

Evidence source

Who vouches for an attribute: attested (signed by the platform under the agent), derived (computed from key material), observed (read from the host), elicited (asked of the agent), or supplied. Kept distinct from the assurance level.

Assurance level

How far a receiver may trust an attribute: self-reported, corroborated or verified. Kept distinct from the evidence source: one names who vouches, the other names how far to trust. Policy grades evidence this way.

Tenant

The customer partition that every agent, token and record belongs to.

MCP bundle

The package you install to try Oathera with your own agent: Oathera Agent Enrollment and Oathera Agent Enforcement.

Bring your own agent (BYOA)

Running Oathera beside an agent you already have, rather than Oathera's sample agent.

Admin API

The authenticated interface operators use to manage the Oathera Identity Service, with roles and separation of duties.

Clearance 0 (Basic)

The baseline clearance: any principal signed in. Every operation requires clearance 0 unless the tenant raises it.

Sovereign (deployment)

Used only to describe a deployment: self-hosted or in-country. Sovereign refers to where Oathera runs, never to a clearance and never as a tagline.

Enrollment review

An administrator's separate sign-off on an enrollment, for separation of duties.

Task context

The orchestrator's signed statement of what an agent is meant to be doing, which bounds its operations.

Delegation

A scoped, signed grant from one agent to a sub-agent, never wider than the delegator's own authority.

Operation approval

A person's signed approval of one specific operation before an agent may perform it. Distinct from enrollment approval, which happens once at enrollment.

Signed revocation snapshots

Signed revocation status, refreshed every 2 seconds for agents at risk tier 1 and up to every 300 seconds at risk tier 4.

Verified sponsor

A person sponsor proofed by one of your proofing sources, or a verified organization. Oathera never proofs anyone itself, and a strong login is not proofing.

Proofing source

An issuer you trust to say a person was proofed. Not every identity provider is one. Oathera relies on your proofing sources rather than verifying people itself.

Clearance 1 (Verified) and Clearance 2 (High)

Clearances above Basic. 1 needs a verified sponsor; 2 also needs a phishing-resistant, hardware-bound sign-in at approval and an attested host. An agent holds the lower of its requested clearance and what the evidence supports.

Handover

A new principal under the same sponsor approving an agent whose principal has left, so it keeps its anchor key.

Attested host

A host whose platform signs a statement bound to the agent's key. On Google Cloud this reaches verified, and a setting makes it mandatory. AWS and Azure documents are checked in part and don't reach verified.

Known organization

An organization sponsor identified by its identity-provider tenant or a verified domain.

Verified organization

An organization sponsor whose legal identity a legal-entity identifier confirms.

Plan

A commercial package of Oathera, which may be measured in sponsor seats. Defined here for vocabulary only; pricing is confirmed at launch.

Agent class

A set of agents sharing one approved definition and operational boundary, so a sponsor can approve them once.

Trust score

A score for each agent derived from its activity and status over time. Not the same as risk tier, assurance level or clearance.

Agent identity wallet

The credentials an agent carries and presents to prove who it is and who stands behind it.

Agentic identity (agent identity, AI agent identity)

The identity an AI agent holds in its own right: bound to the machine it runs on, with a named person who answers for it, and checked every time it acts. Also called AI agent identity, autonomous agent identity, digital actor identity or first-class agent identity. Read more.

Agentic identity management (AIM)

Enrolling, approving, checking and revoking the identities of AI agents across their life. Read more.

KYA verification

Checking, before an AI agent acts, which agent it is, which person answers for it and whether it may do this. Know Your Agent is the category; Oathera is the company and product. Read more.

NHI governance and lifecycle management

Inventorying, approving, monitoring and retiring non-human identities such as service accounts, keys, certificates and AI agents. Read more.

Non-person entity (NPE)

A term used in US federal and defence settings for an identity that is not a person. It maps to non-human identity. Read more.

Machine identity

Usually the certificates and keys that identify devices and workloads. An AI agent is a machine identity that also makes decisions, so it also needs an accountable person and a check on each action. Read more.

Agentic IAM

Identity and access management built for AI agents: per-agent identity, a named accountable person, authorization on every call and governance over the agent's life. Read more.

Enterprise AI agent governance

The programme around agent controls: who may enroll an agent, who approves it, what it may reach, how it is reviewed and how it is stopped. Read more.

Accountable sponsor (principal-to-agent binding)

The person or organisation that answers for an agent, recorded with its enrollment. The principal's approval binds the agent to that person. Read more.

Non-exportable anchor key

The agent's long-lived Ed25519 key, created and kept on its host. It can be held in KMS, HSM, TPM or a secure enclave so the key material is not exported. Its key digest is the agent's identity. Read more.

Proof of possession for agents (sender-constrained identity)

A token is useless without the key it names. Oathera uses an anchor key, a rotating in-memory session key and a single-use operation proof for each call. Also searched as sender-constrained agent identity. Read more.

Key-binding JWT

A single-use JWT signed by the anchor key that names the new session key at each token issuance. Its type is oathera-auth-keybind+jwt and it lasts 120 seconds. Read more.

Observed vs asserted attributes

What Oathera finds on the host (observed) versus what the agent says about itself (asserted). Evidence is graded as self-reported, corroborated or verified, and clearance follows the evidence. Read more.

Agent provenance (audit trail)

Evidence of which agent performed an action and under whose authority, that can be re-verified later. Held in a hash-chained audit log with signed seals. Read more.

Agent-to-agent (A2A) authentication

One agent verifying another agent's identity from fresh evidence outside the message text, instead of trusting a name written in a message. Read more.

Agent revocation and kill switch

Revoking an agent refuses its next token request and next boundary decision. Tokens already issued stay valid until they expire, five minutes at most. A kill switch suspends an agent and reinstates it after its host is checked again. Read more.

MCP security

Securing Model Context Protocol servers: knowing which agent is calling, whether a person answers for it and whether this exact call is allowed, checked on every request in front of the tool. Read more.

Confused deputy

An agent that holds authority a caller lacks and uses it on that caller's behalf, often after a planted instruction in content the agent read. Read more.

Prompt-injection privilege escalation

An attacker uses injected instructions to make an agent use access it already holds. Limited by narrow boundaries and checks on every call, outside the model. Oathera does not scan content for injected instructions. Read more.

Token theft and replay

Using a copied token, or a captured proof, again. Single-use operation proofs, a token that names its session key and host binding stop it. Read more.

Shared API key

One secret used by many agents. It hides which agent acted and cannot be revoked per agent. Read more.

Shadow AI agents

Agents running without enrollment or approval. Found at the tool: a call with no identity is an unenrolled agent. Read more.

Blast radius

How much damage one compromised or misled agent can do with the access it holds. Read more.

Over-privileged service account

A shared account that has collected more permissions than any one task needs. An agent running on it inherits all of them. Read more.

Just-in-time authorization

Authority granted for the operation at hand and not held in advance. Each enforced call needs a signed, single-use boundary decision. Read more.

Delegation chain attribution

Showing which person authorized which agent, and which agent delegated to which, for every action in a multi-agent chain. Read more.

EU AI Act accountability

The EU AI Act expects record-keeping and human oversight for high-risk systems. Oathera's identity, approval and decision records can support that. They do not make a system compliant. Read more.

DPoP (Demonstrating Proof-of-Possession)

A standard for binding OAuth tokens to a key. Oathera does not use DPoP: the single-use operation proof plays that role for each call and covers its arguments. Read more.

SPIFFE and SPIRE

Standards and a runtime for workload identity. A SPIRE-issued JWT-SVID can be optional enrollment evidence for an allow-listed trust domain; the agent's identity stays its anchor key. Read more.

Verifiable credentials and DIDs

The credential service issues W3C Verifiable Credentials recording who approved what. Decentralized identifiers are not how Oathera names an agent: the identity is the key digest. Read more.

← Back to Oathera
Oathera logo Oathera

Every AI agent, sworn in. Oathera gives every AI agent an identity bound to the machine it runs on, with a named person who answers for it, and checked every time it acts.

Product

  • Use cases
  • Platform
  • Features
  • Integrations
  • Security

Developers

  • Developer section
  • Data sheets
  • Learn
  • Glossary
  • Simulator

Company

  • About us
  • Blog
  • Careers
  • Contact
  • GitHub

Legal

  • Privacy Policy
  • Terms of Service
  • Data Processing Agreement
  • Sub-processors
© 2026 Oathera Inc. · All rights reserved
Know Your Agent (KYA)

Cookie preferences

We use cookies to run this site and, with your consent, to understand usage and improve Oathera. Strictly necessary cookies are always on; you can choose whether to allow analytics and marketing cookies below.

Read the privacy policy

  • Strictly necessaryAlways on

    Required for the site to work, including security, load balancing, and remembering your cookie choices. These cannot be switched off.

  • Help us measure traffic and see how the site is used, so we can improve it. No personal profiles are built.

  • Used to make messages about Oathera more relevant across other sites. Off unless you turn it on.