All chapters
  1. Start hereA valid token is not a permission
  2. Foundations · 1Identity and trust primitives: who is calling, and who checks
  3. Foundations · 2People and identity proofing: accounts, passwords and recovery
  4. Foundations · 3Sessions and browser security: cookies, CSRF, XSS and the BFF
  5. Foundations · 4Authenticators and MFA: passwords, passkeys and phishing resistance
  6. Foundations · 5API keys and application credentials: what a key proves
  7. Foundations · 6Tokens: bearer, JWT anatomy, validation and revocation
  8. Foundations · 7OAuth, OpenID Connect and federation: delegation and login
  9. Authorization · 8Authorization from first principles: subject, action, resource, context
  10. Authorization · 9Access control models: RBAC, ABAC, ReBAC and ownership
  11. Authorization · 10The enforcement architecture: PEP, PDP, PIP and PAP
  12. Authorization · 11Policy as code: versions, tests, engines and decision logs
  13. Authorization · 12Tenant and data isolation: one platform, many brokerages
  14. Authorization · 13Identity governance and lifecycle: joiner, mover, leaver
  15. Authorization · 14Privileged access: standing rights, just-in-time elevation and break-glass
  16. Workloads · 15TLS, PKI and certificates: proving a host, and a client
  17. Workloads · 16Workload and service identity: SPIFFE, mTLS and caller allowlists
  18. Workloads · 17API gateways and zero trust: the edge checks, the service decides
  19. Workloads · 18Event-driven identity: Kafka principals, ACLs and replay
  20. Workloads · 19Cloud IAM and secretless deployment: roles, federation and no stored keys
  21. Workloads · 20External SaaS, webhooks and trust: calls that cross company lines
  22. Workloads · 21Private controlled environments: what private networking does and does not secure
  23. AI-native · 22Authenticating to model providers: keys, gateways and cloud IAM
  24. AI-native · 23MCP identity and authorization: discovery is not permission
  25. AI-native · 24RAG and retrieval authorization: only what this caller may read
  26. AI-native · 25Agents as principals and delegates: whose authority is this?
  27. AI-native · 26Agent-to-agent trust: identity, capabilities and forwarded authority
  28. AI-native · 27Coding agents and CI: repository scopes, OIDC and no production keys
  29. Operations · 28Auditing and observability of access: decisions you can prove later
  30. Operations · 29Threat modeling and testing: the negative tests that prove access control
  31. Operations · 30Operational lifecycles: rotation, revocation and outages
  32. Reference · 31Standards, frameworks and products: what each one is, and what it is not
  33. Reference · 32Architecture decision playbooks: choosing an auth strategy
  34. ReferenceGlossary
  35. AboutAbout this site and its sources
Reference

Glossary

Every term this site uses, each with what it is, what it is not, and where it shows up in the insurance example.

Every term this site uses, in one place. Each card opens the same drill-down as the underlined phrases in the text: one line, what it is, what it is not, and how it shows up in the insurance example.

Two sentences hold the words apart. A valid token is not a permission. And: the model recommends; a person decides. Authentication proves who is calling and how sure we are. Authorization decides whether this caller may do this action to this record, now. Enforcement makes the decision stick on every path, and audit records it without the secrets.

A2AOne agent asking another agent, often in another organization, to do work; also the open protocol (specification 1.0.0) for publishing agents, sending messages and tracking tasks.AALHow robust a login is: how confident the service is that the person controls the account's authenticators right now.ABACA model where attributes of the subject, the resource, the action and the environment are evaluated against policy.Access reviewA periodic decision, by someone who knows the work, to keep or remove each grant.Access tokenA token a client presents to a resource server to call it, scoped and audience-restricted.Access-removal targetThe maximum time an organization accepts between deciding to remove access and the last request that succeeds with it.Account lifecycleThe life of an account: created, granted attributes, changed when the person moves, suspended, and removed.Account lockoutA limit on failed sign-in attempts for one account, by blocking it for a time or by adding growing delays.Account recoveryThe process that lets a person back in after losing a password, phone or key, without giving attackers a cheaper login.ACLA list stored with a resource that names principals and the operations each may perform on it.Actor claimThe JWT claim, defined in RFC 8693, that names the party acting on behalf of the token's subject.Agent CardA JSON document an A2A agent's provider publishes with the agent's identity, skills, endpoint and authentication requirements.Agent principalAn agent treated as a principal in its own right, with a workload identity and permissions granted to it as an application.Agent sandboxA container or virtual machine where the agent edits files and runs tests, with only the network access and credentials the task needs.ANN recallThe share of the true nearest neighbours that an approximate nearest-neighbour search actually returns.API gatewayThe single entry point that routes API calls and applies cross-cutting checks: token validity, audience, route scope, rate limits.API keyA string a service issues to a calling application or account; whoever presents it is treated as that caller, and nothing more is proven.Architecture decision recordA short document, kept with the code, that records the context, the decision, the alternatives and the consequences of one design choice.AssuranceHow strong the evidence was that the claimant controls the authenticators of this identity, at that moment.AudienceThe claim that names the service or services meant to accept the token.Audit integrityThe property that a record, once written, cannot be changed or removed without the change being detected.Audit recordA record that lets a person reconstruct who did what to which resource, under which policy, with what result.Audit trailThe ordered, append-only, access-controlled collection of records saying who did what to which record, under which policy, with what result.AuthenticationVerifying proof that a caller is who it claims to be, to some level of confidence.AuthenticatorThe thing a person uses to demonstrate control or possession of an authentication factor.AuthorizationDeciding whether this principal may perform this action on this resource, in this context, now.Authorization codeThe grant in which the browser brings back a one-time code that the client's backend exchanges for tokens over a direct call.Authorization policy versionA release name for one exact text of the authorization rules, stamped into the policy and into every decision record.Authorization serverThe OAuth role that authenticates the resource owner or client and issues access tokens.Backend for frontendA server component that belongs to one front end, acts as its confidential OAuth client, keeps its tokens, and proxies its calls so the browser never sees a token.Bearer tokenA token that works for any party in possession of it, with no further proof.BFLAA caller reaches a function, often an administrative or write operation, that their role should never call.BOLAA valid, authenticated caller reaches a specific record because nobody checked that record against the caller.Branch protectionRepository rules (classic branch protection or rulesets) that require pull requests, approvals, checks or signatures before a branch changes.Break-glassPrivileged access granted at once in an emergency, for a fixed short time, with an immediate alert and a mandatory review by someone else.BrokerA person at a brokerage firm who asks the insurer for quotes through the broker portal.Budget enforcementA limit on spend, requests or tokens that blocks further calls once reached, per application, workspace or account.Business-operation checkAn authorization check written for the business operation (record a decision on this submission) rather than the endpoint (POST a URL).Caller allowlistThe receiving service's own list mapping each operation to the exact verified workload ids allowed to call it, denying everything else.CedarA language and engine where permit and forbid policies over typed principals, actions and resources decide Allow or Deny.Certificate authorityA party whose signature on a certificate the verifier has chosen to trust.Certificate chainThe ordered list of certificates from the leaf up to a root the verifier trusts, each signed by the next.Certificate revocationTelling verifiers that a certificate must no longer be accepted, usually because its key leaked.Certificate rotationSwitching a workload to a new key and certificate while the old one is still valid, then retiring the old one.Change managementThe process by which a change is proposed, reviewed by someone other than its author, approved, recorded and released.Chunk ACLThe access information a chunk carries from its source document, so a search can decide on each chunk without asking the source first.CI OIDC federationThe CI system issues a signed OIDC token per job, and the cloud issues short-lived credentials only if the token's claims match its conditions.ClaimA name-value statement in a token, such as sub, aud or tenant.Client credentialsThe grant in which a client requests an access token with only its own credentials, acting for no person.Client secretA secret shared between an application and an authorization server, sent to prove the application holds its client_id.client_idThe identifier an OAuth authorization server assigns to an application; it names the client and proves nothing.Complete mediationEvery access to a protected resource goes through the authorization decision, on every path, each time.Confused deputyA program with authority that is tricked into using it for someone who lacks that authority.ConsentThe resource owner's agreement that a particular client may receive particular scopes on their behalf.Consumer groupA named set of consumers that split a topic's partitions among themselves and commit their read positions together.Correlation idAn identifier carried across the records of different services and times so they can be joined into one story.CORSA set of HTTP headers by which a server tells browsers which other origins' scripts may read its responses.CredentialSomething a principal holds and presents as proof: a password, a private key, a client secret, a token.Cross-organization federationAn arrangement in which one organization's authorization server accepts identity or authorization information issued by another's, under agreed rules, and issues its own token.Cross-site request forgeryAn attack in which a page on another site makes the victim's browser send a request that carries the victim's automatically attached credentials.Cross-site scriptingAn attack in which the attacker's script runs inside a trusted page, with that page's origin and privileges.Cross-tenant testA test that acts for one tenant with entirely valid credentials, asks for another tenant's data on one path, and passes only when nothing comes back.Data residencyA constraint, from a contract or a law, on the jurisdiction or region where a tenant's data is stored and processed.Database per tenantAn isolation model where each tenant's data lives in a separate database, reached with separate credentials.Dead letter topicA topic where a consumer parks events it could not process, so the main flow keeps moving and the failures can be examined and replayed.Decision cacheA store of recent PDP answers keyed by the request, reused instead of asking again.Decision logOne record per authorization decision: who asked, for which action and resource, the answer, the rule, the reason and the policy version.Decision playbookA worked decision for one kind of system: the principals, the simplest credible option, the trigger to grow, the alternatives, the failure modes and a checklist.DecisionThe result of evaluating policy for one request: allow or deny, with a status, the deciding rule and a reason.Defense in depthSeveral independent controls in a row, each able to stop a failure the others miss.Delegation chainThe sequence of actors through which delegated authority passed, recorded as nested act claims.DelegationA principal acting with authority another principal granted it, narrowed to a task, while keeping its own identity.Deny by defaultAccess is refused unless a rule explicitly allows it.Device flowA grant for devices without a usable browser: the device shows a short code, the person approves it on a phone or laptop, and the device polls for the token.Device postureFacts about the machine or workload making a request, such as managed or not, patched, encrypted, security agent running, that a policy can use.Discovery documentA JSON document at a well-known URL that lists an identity provider's issuer, endpoints, keys and supported features.Down-scopingIssuing a token with fewer scopes, a narrower audience or a shorter life than the token it was derived from.Dual controlA privileged action needs one person who asks and a different person who agrees, at that moment.Egress controlA proxy or firewall rule set that lets workloads connect only to a named list of outside destinations.Emergency revocationEnding a credential's or a person's access early: a stolen token, a compromised key, a leaver who must be out now.Enforcement pointThe code in the request path that asks for a decision and applies it before any data moves.Enterprise identity providerThe system that authenticates the organization's people (and often workloads) and issues their tokens, fed by the HR source of truth.Enterprise-managed authorizationAn official MCP extension that lets an organization control MCP server access centrally through its existing identity provider.EntitlementA specific permission granted to one principal, often for a class of resources or one resource.Ephemeral build identityA credential created for one CI job and expired with it, instead of a standing service account or stored key.Escalation triggerThe observable change that shows the assumption behind a simple design no longer holds, so the design must grow.Event provenanceFacts inside an event that say who or what caused it and where it came from, written as identifiers rather than credentials.Fail closedA design where an error, timeout or missing answer in a security check leads to denial, never to permission.Fail openA control that lets a request through when its decision point, key set or identity provider is unavailable or errors.FederationAn agreement in which one organization asserts identities and another relies on those assertions, across a boundary neither controls alone.FIDO2The FIDO Alliance name for the pair WebAuthn (browser and platform API) and CTAP2 (external authenticators over USB, NFC, Bluetooth).Field-level securityControls that hide or refuse individual fields of a record the caller may otherwise read or write.Filter-firstApplying the authorization constraint as part of the search, so only chunks the caller may read are candidates and ranking happens among them.Forwarded authorityHow a person's or upstream principal's authority reaches a recipient further down a chain; done by obtaining a new token for that recipient, never by forwarding the one received.GitHub AppAn integration registered with GitHub that acts as itself, with fine-grained permissions, on the repositories its installer chose.HTTP 401The request lacks valid authentication credentials, so the server does not know or does not trust who is calling.HTTP 403The server knows who is calling and refuses the action; 404 instead when the caller should not learn that the record exists.HttpOnlyA Set-Cookie attribute that forbids JavaScript from reading the cookie, while the browser still sends it on requests.Human approvalA separate, authorized request by a person, through a client built for that person, before an irreversible effect happens.IALHow robust the identity proofing was: how sure we are that an account belongs to a particular real person.IAM roleAn AWS identity that trusted principals take on to receive temporary credentials; it has a trust policy (who) and a permissions policy (what).ID tokenAn OpenID Connect token, always a JWT, that tells the client application who authenticated, when and how.IdentificationA caller says who it claims to be, before anything is proven.Identity proofingChecking evidence (a document, an authority's record, a live session) to link an account to a specific real-life person.Identity propagationPassing the identity of the person (or upstream principal) a workload acts for to the next service, alongside the workload's own identity.Identity providerSoftware or a hosted service that authenticates people or machines and issues tokens or assertions about them, usually by OAuth 2.0, OpenID Connect or SAML.Identity provider outageA period when the identity provider cannot authenticate anyone or issue tokens.IdentityThe set of attributes that distinguishes one principal from every other in a given system.IDORA caller changes an identifier in a request and reaches an object the server never checked against the caller.Incident responseThe planned sequence for a security incident in access controls: contain the credential, rotate what may be exposed, investigate by correlation ids, repair what was too broad.Index deletionRemoving a source document's chunks and vectors from every search structure, and proving it happened.Internal certificate authorityA certificate authority the organization runs to issue certificates for its own hosts, workloads and clients.IntrospectionA resource server sends a token to the authorization server and gets back whether it is active and, if so, its claims.IssuerThe claim naming the authority that issued the token, matched exactly against the expected value.Joiner, mover, leaverThe three events in a person's working life that change access: arriving, changing role or unit, and leaving.Just-in-time elevationAn eligible account requests privileged access with a reason, a second person approves, and the access expires by itself.Just-in-time provisioningThe application creates the account the first time a person signs in through federation, from the assertion's attributes.JWEAn encrypted and integrity-protected message, five base64url parts in compact form.JWKSA JSON document listing an issuer's public keys, each with a kid, so verifiers can check signatures and follow rotation.JWSA signed or MACed message in three base64url parts: protected header, payload, signature.JWTA token format: a JSON object of claims encoded in a JWS (signed) or a JWE (encrypted).Kafka ACLA rule that allows or denies one operation to one principal from a host on resources matching a literal or prefixed pattern.Kafka principalThe identity Kafka assigns to a connection after authentication, written PrincipalType:name, such as User:intake-producer.Key restrictionLimits attached to a key: which APIs it may call, from which networks or apps, for which workspace, until when.Key rotationIssuing a new key, moving every user to it while both work, then revoking the old one.KeycloakAn identity and access management server that manages realms of users, clients, roles and groups and issues OpenID Connect and SAML tokens.Least privilegeEach principal gets only the permissions its job needs, for as long as it needs them.LLM gatewayAn internal service every application calls instead of the provider; it authenticates each application, enforces model policy and budgets, logs, and holds the only provider credential.Log redactionRemoving or masking sensitive values (tokens, passwords, keys, session ids, user content) before a record is written.Managed identityAn Azure identity whose credentials the platform handles, so code gets Entra tokens without any secret.Maximum authorityThe rule that an agent never exceeds the intersection of its own ceiling, the delegator's grant and the task.MCP serverThe program that offers tools, resources and prompts over MCP and runs a tool when a client calls it.MCPAn open protocol through which an AI application (the host) connects to servers that offer tools, resources and prompts.Message signatureA signature over a message's content with the sender's key, verifiable later wherever the message has been, unlike TLS, which protects only the connection.Metadata serviceAn HTTP endpoint reachable only from inside a machine that describes the instance and returns temporary credentials for its attached identity.MFAA login that needs evidence from at least two different kinds of factor: something you know, have, or are.Model permissionThe rule that says which application or principal may invoke which model, enforced by the gateway or by cloud policy.mTLSTLS in which the client also presents a certificate and proves it holds the private key, so both sides know who is on the other end.Negative testA test that sends a request which must be refused, usually with otherwise valid credentials, and passes only when it is refused and nothing changed.OAuthA framework in which a client obtains a limited, revocable access token from an authorization server that the resource server trusts.OAUTHBEARERA SASL mechanism in which the client presents an OAuth access token at login and the broker validates it.OPAAn engine that evaluates policies written in Rego against structured input and data and returns a decision document.Opaque tokenA random-looking token that carries no claims; the resource server learns them by asking the issuer.OpenFGAAn engine in the style of Google's Zanzibar that answers whether a user has a relation to an object, from stored relationship tuples.OpenID ConnectAn identity layer on OAuth that returns an ID token telling the client who authenticated, when and how.Orphan accountAn active account that no current person or team answers for.Outbound integrationYour service calling a vendor's API, with a credential the vendor issued and limits you set.OwnershipA relationship that gives the owning principal or organization rights over a record.PAMThe practice and tooling that controls access able to change who may do what: who is eligible, how it is activated, for how long, approved by whom, and what record remains.PAPThe place where authorization policy is authored, reviewed, approved and released as a version.PasskeyA key pair created for one site and kept in an authenticator; the site stores the public key and checks signatures over fresh challenges.Password hashThe stored output of a deliberately slow, memory-hungry one-way function, so a leaked table does not reveal passwords.PDPThe code that evaluates policy for a subject, action, resource and context and returns allow or deny with a reason.PEPThe code in the request path that holds a request, asks for an authorization decision, and enforces the answer.Personal access tokenA GitHub token a person creates for scripts and tools; whatever uses it acts with that person's access.Phishing resistanceThe property that an authentication protocol does not disclose a reusable proof to a site pretending to be the real one.PIMThe identity-provider side of privileged access: eligible role assignments that a person activates for a limited time.PIPA system that supplies attribute values a decision needs and the request does not carry.PKCEThe client sends a hash of a secret with the authorization request and the secret itself at the token request, so a stolen code is useless.Policy as codeAuthorization rules kept as files in a language made for rules, in version control, reviewed, tested and released under a version name.Policy cache invalidationHow a change in policy or in the facts it reads reaches authorization answers that services have cached.Policy driftThe state where some decision point evaluates a different policy version than the released one.Policy engineA component that evaluates policy in its own language against the facts it is given and returns a decision, such as OPA with Rego, Cedar or OpenFGA.PolicyThe rules that turn subject, action, resource and context into allow or deny.Post-filterFinding the nearest chunks without regard to permissions, then removing the ones the caller may not read.PrincipalWhoever a request is made by or for: a person, an application, a workload, a service or an agent.Private endpointA network interface inside your private network that reaches a cloud service without crossing the public internet.Privilege escalationA principal ends up able to do something beyond its grants, by asking for more, by a misconfiguration, or by a policy that checks too little.Prompt loggingKeeping the text of model requests and responses, at the gateway, in traces or at the provider.Proof of possessionA token bound to a key the legitimate client holds, so the caller must also prove it holds that key on each request.Protected resource metadataA small JSON document in which a protected resource says what it is and which authorization servers issue tokens for it.Provider API keyA key a model provider issues that identifies an account, a person or a service account on every model request.ProvisioningCreating, changing and deactivating accounts and group memberships in the systems that need them, driven by an authoritative source.QuoteThe record a broker sees, identified by a quote_id the portal backend creates; it maps one to one to an internal submission.RBACA model where a principal's roles decide which actions it may perform.ReBACA model where links between principals and objects (assigned, member, owner, shared with) decide access.RecommendationThe underwriting agent's suggested outcome for a submission, with reasons and citations, stored for the underwriter to review.Refresh tokenA long-lived credential a client sends only to the authorization server to obtain new access tokens.RegoOPA's declarative language: rules that become true when every expression in their body is true.Replay protectionChecks that stop a captured or resent message from being accepted twice: a signed timestamp within a window, and remembered message ids.Request idAn identifier created once per inbound HTTP request at the edge and passed to every service that handles it.Residual riskThe risk that remains after a control is in place and working, written next to the control so it gets an owner.Resource indicatorThe resource parameter with which a client tells the authorization server which protected resource a token is for, so the token's audience names it.Resource serverThe OAuth role that hosts protected resources and accepts access tokens issued for its own audience.Retrieval authorizationDeciding, for each chunk a search could return, whether this caller may read it now, and making sure only those chunks reach the model.RevalidationRechecking each search result against the source system's current access rule at read time, before the text reaches the model.Role hierarchyAn arrangement where a senior role automatically includes every permission of the roles beneath it.Row-level securityA database feature that limits which rows a session can read or write, according to policies on the table.Run idAn identifier for one execution of the agent workflow for a submission, carried on every request the agent makes during it.Runner secretA secret stored in the CI system at repository, organization or environment level and handed to jobs at run time.SaltA unique random value fed into the password hash with the password, so equal passwords get different hashes.SameSiteA Set-Cookie attribute (Strict, Lax or None) that decides whether the browser attaches the cookie to requests started by another site.SAMLAn OASIS standard in which an identity provider sends signed XML assertions about a user to a service provider, usually through the browser.SANThe certificate extension that carries the names a certificate is valid for: DNS names, URIs, IP addresses.SASLSimple Authentication and Security Layer: lets a protocol such as Kafka offer several authentication mechanisms through one interface.Schema registryA service that stores the schemas of a topic's events by version, so producers and consumers agree on the format.SCIMAn HTTP protocol and schema for creating, changing and deactivating users and groups across systems.ScopeA coarse permission name requested by a client and carried in its access token, such as quotes.read.SCRAMSalted Challenge Response Authentication Mechanism: the client proves it knows a password through a challenge, without sending it.Secret leakageA key, password or token copied to a place where people or systems that should not hold it can read it.Secrets managerA service that holds secrets such as vendor API keys, checks the caller's identity and permission on each read, records the read and supports rotation.Security eventA record that something security-relevant happened (a failed sign-in, a denial, a revocation, a key change), written so a person or a rule can react now.Security schemeAn Agent Card's declaration of an accepted authentication method, one of five types based on the OpenAPI 3.2 Security Scheme Object.Separation of dutiesThe requester, the approver and the beneficiary of access are different people, and incompatible actions are never held by one principal.Service accountA non-human identity that a program runs as; on Google Cloud, the identity attached to compute or impersonated through federation.Service meshInfrastructure that puts a proxy beside each workload to run mutual TLS between them and enforce coarse rules about which workload may call which.Session cookieA cookie whose value is an opaque, random id that points to a session record kept on the server.Session fixationAn attack in which the victim logs in under a session id the attacker already knows, because the server kept the id across login.Session recordingCapturing a privileged session, typically a remote shell or desktop connection, for later review.SessionState a server keeps so that a person who authenticated once is recognized on later requests.Shared schemaAn isolation model where every tenant's rows live in the same tables and a tenant column plus a predicate or policy separates them.SIEMA system that collects security events from many sources, stores them for search, and runs detection rules that raise alerts.Source of truthThe system whose record governance trusts for a fact, usually HR for employment and unit.SPIFFE IDA URI of the form spiffe://<trust domain>/<workload identifier> that uniquely names one workload.SPIFFEA set of open standards that define a workload's name (the SPIFFE ID), the documents that carry it (SVIDs), the local API that delivers them, and federation between trust domains.SPIREThe reference implementation of the SPIFFE APIs: a server that signs SVIDs and agents on each node that attest workloads and serve the Workload API.Spring SecurityA Java library that provides a security filter chain, an OAuth 2.0 resource server for JWTs, and method security such as @PreAuthorize through AuthorizationManager.StandardA document from a standards body that says what implementations MUST and SHOULD do, such as an RFC, an OpenID specification or a W3C Recommendation.Standing privilegePrivileged access held all the time, with no request, no approval and no expiry.Stdio trustThe trust model of a local MCP server that talks over standard input and output: no token, credentials from the environment, the operating system account as the boundary.Step-upA server refuses a sensitive action until the person authenticates again at a higher level, then lets the same session continue.STRIDEA threat classification: Spoofing, Tampering, Repudiation, Information disclosure, Denial of service, Elevation of privilege.STSA token service that exchanges proof of identity for credentials that expire.SubjectThe principal a decision is about, with the verified attributes the policy reads.SubmissionThe insurer's internal record of a quote request, about 63 fields in the scenario, worked by underwriters and the underwriting agent.Supply-chain provenanceA signed statement that an artifact was built from a given source, by a given workflow, on a given platform.SVIDThe document that carries exactly one SPIFFE ID: an X.509 certificate (X.509-SVID) or a signed JWT (JWT-SVID).Task tokenA short-lived delegated token bound to one run, one record, a list of actions and an expiry.Tenant isolationThe guarantee that no principal acting for one tenant can read, change or infer another tenant's data through any path.TenantOne external customer organization that shares an application and its data stores with other customers.Third-party access reviewA periodic review, by a named owner, of each outside company's access and the data sent to it, ending in keep, narrow or remove.Threat modelA written list of the system's trust boundaries, the threats at each, the control for each threat and how that control is verified.TLSThe protocol that encrypts a connection and authenticates the server, and optionally the client, with certificates.TOCTOUA race in which access is checked, the facts change, and the action then runs on the old answer.Token exchangeA client presents a token it holds (and optionally its own as the actor) and asks the authorization server for a new token for one audience with equal or fewer scopes.Token formatA standard that defines the structure of a token and how its integrity or confidentiality is protected, such as JWT over JWS or JWE.Token passthroughAn MCP server forwarding a token it received from a client to another service, instead of obtaining a separate token for that service.Token revocationThe issuer marks a token as no longer usable before its exp; only resource servers that check learn of it.Token substitutionA genuine token issued for one recipient or purpose is presented at another, hoping it is accepted there.Tool discoveryThe step in which a client asks an MCP server which tools exist, and receives names, descriptions and argument schemas.TOTPA short code that the app and the server both compute from a shared secret and the current 30-second time step.Trace idA 32-hex identifier for one trace; on this site one trace covers one inbound request through every hop, with a 16-hex span id per hop.Trust boundaryA line where a request or data passes from a part you have not verified into a part that will act on it.UnderwriterThe insurer's employee who reviews the evidence and the agent's recommendation and decides the submission in the underwriter cockpit.Verified emailAn email address the account owner has shown they can read, by returning a code or clicking a link sent to it.WAFA filter that inspects HTTP requests for known attack shapes, such as injection payloads, malformed headers and abusive clients.WebAuthnThe browser API through which a site asks an authenticator to create a public key credential and later sign challenges with it.Webhook signatureA header value the sender computes over the delivery with a secret or private key, which the receiver recomputes or verifies before trusting anything in it.WebhookAn inbound HTTP request, usually a POST, that tells you something happened in another system.Workload identity federationA cloud trusts tokens from another identity provider and exchanges them for short-lived cloud credentials, so outside workloads need no stored key.Workload identityThe name a program proves when it calls another program, from a credential it holds and the receiver can verify.x-app-idA request header in which a caller names its own application; any program can type it.X.509 certificateA CA-signed statement that a public key belongs to a name for a stated time, profiled for the internet by RFC 5280.Zero trustAn approach in which no network location or earlier login grants trust, and every access to a resource is decided dynamically and enforced before it happens.Zero-downtime rotationReplacing a signing key, secret or certificate through an overlap in which both old and new are accepted, in an order that breaks nothing.

Each term is explained in full on one page and used everywhere else. Statements about standards, protocols and named products were read in their own documentation on 2026-11-18; the sources are listed on About. For the decisions these terms support, see Playbooks.