Cybersecurity · Nov 2026
Machine Identity Inventory: The Missing Security Register
Service accounts and workload credentials can outlive the systems that created them. This guide explains how to assign owners, control rotation and verify retirement.
A machine identity inventory is a register of the accounts, certificates, keys and workload credentials that let software act without a person signing in. It is not a certificate spreadsheet alone. An expired certificate can stop a payment service, while an abandoned API key can preserve privileged access after its owner leaves. These are distinct risks and require different controls. A useful inventory therefore connects each credential to a service, an accountable person and an operational consequence.
The invisible access layer
A machine identity inventory is a register of the accounts, certificates, keys and workload credentials that let software act without a person signing in. It is not a certificate spreadsheet alone. An expired certificate can stop a payment service, while an abandoned API key can preserve privileged access after its owner leaves. These are distinct risks and require different controls. A useful inventory therefore connects each credential to a service, an accountable person and an operational consequence.
Discovery should cover cloud identity platforms, certificate authorities, secret stores, source repositories, CI pipelines and runtime configuration. Compare issued credentials with actual usage. A record that exists only in a vault may be unused; a token embedded in an old integration may never appear in the central inventory. Treat the difference as an investigation queue rather than assuming either source is complete.
Ownership is a control, not a label
Record the business service, technical custodian, accountable owner, issuer, permission scope, environment, expiry, rotation method and incident contact for every important identity. Team aliases are useful routing tools but cannot replace a person empowered to approve rotation or revocation. When a service changes teams, machine identities must move with its architecture diagrams, runbooks and support commitments.
Define ownership exceptions explicitly. A privileged credential without a current owner should trigger escalation and temporary monitoring, not immediate deletion that could interrupt a critical service. The decision belongs to the business risk owner with security advice and an expiry date for the exception.
First discover, then enforce
During days zero to thirty, focus on mapping and classification. Separate production credentials from test artifacts, persistent service accounts from short-lived workload identities, and automated rotation from manual change windows. A broad campaign to disable every old-looking identity can create a major outage without proving that the exposure has disappeared.
Prioritise by privilege, internet exposure, age, criticality, evidence of use and ease of replacement. This is a decision aid, not a scientific probability of compromise. Investigate highly privileged credentials that have no owner and no expiry first. Keep an audit trail of what was found, who reviewed it and why action was deferred.
Rotation must include consumers
Where supported, prefer workload federation and short-lived credentials instead of static secrets. When a persistent secret remains necessary, map its consumers and establish a tested dual-key transition. Rotate in staging, confirm every authorised integration can authenticate, and only then withdraw the previous key in production.
An apparently successful rotation may conceal an application that still uses cached credentials. Verify both positive and negative cases: the new identity performs approved operations and the old one can no longer authenticate after the overlap period. Keep the recovery plan available if a dependency was missed.
Revocation needs independent evidence
Closing an access ticket is not the same as ending access. Check issuer state, downstream permissions, active sessions where technically revocable and the maximum remaining validity of issued tokens. A signed token may remain usable until expiry even after its issuing account is disabled; this is one reason to minimise lifetime and scope.
Include service accounts and automation tokens in contractor offboarding. The related article on third-party access offboarding explains the need for closure evidence. Capture who tested the denial, when the test occurred and whether legitimate service traffic continued.
Board metrics should expose gaps
Useful measures include coverage of known issuers, percentage of privileged identities with named owners, overdue rotations, orphaned credentials, use of short-lived authentication and median time to revoke. Always show the inventory denominator. Ninety percent compliance across half of the estate is not equivalent to ninety percent of the entire estate.
Report major exceptions with a service impact statement, owner, compensating control and deadline. Separate outage risk from unauthorised access risk, since the right response differs. The goal is a decision-ready register, not another dashboard whose green colour hides uncertainty.
A hypothetical migration
Consider a hypothetical payroll integration using one static API key across development, staging and production. The team introduces environment-specific identities with narrower privileges, migrates consumers in stages and monitors successful transactions. Only after validating the new path does it revoke the shared key.
This is an illustrative scenario, not a claim about a client engagement. The evidence consists of failed authentication with the old key, successful authorised transactions and sign-off from the service owner. Without those tests, a migration may simply move the risk to another location.
Ninety days to an operating control
Days zero to thirty: discover issuers, build the register, assign ownership and document high-risk exceptions. Days thirty-one to sixty: reduce privilege, automate rotation for selected services and add credential checks to deployment reviews. Days sixty-one to ninety: perform revocation exercises, audit ownership changes and report unresolved risks to leadership.
Machine identities also underpin autonomous tooling. The article on agentic AI permissions extends the same ownership principle to tools that take actions. Reference points include NIST SP 800-207 and CIS Controls. If you do one thing this month, identify the owner of every privileged non-human identity before changing its credentials.