Back to Blog
Access Management
8 min read

Privileged Access Management: Vaults, Secrets and Machine Identities

Most PAM programmes secure human administrators and ignore the service accounts, which outnumber them heavily and never change their passwords.

GuardsArm Team

Security Experts

May 19, 2025

Privileged access management

PAM programmes usually begin and end with human administrators — vault the domain admin passwords, enforce check-out, record sessions. That work matters, and it addresses a minority of the privileged credentials in the estate.

The majority are machine identities: service accounts running applications, integration credentials, API keys, certificates, database connection strings. They typically have high privilege, passwords that have never changed, and no owner. The human side is covered in privileged access for clinicians; this is the other half.

Outnumber humans
Service and machine identities typically exceed human admins by an order of magnitude
Never rotated
Changing them risks breaking something nobody fully understands
Often unowned
Created during a project, inherited by nobody

Find them first

Service accounts hide because they were created outside normal process:

Where to lookWhat you find
Directory accounts with non-expiring passwordsThe classic service account signature
Accounts with a service principal nameKerberos-authenticated services
Scheduled tasks and service definitionsThe account each runs as
Application configuration filesConnection strings and embedded credentials
Source repositoriesCommitted secrets, including in history
CI/CD pipeline variablesDeployment credentials, often highly privileged
Certificate storesExpiry dates nobody tracks

Repository history is worth emphasising. A secret removed in a later commit is still in the history and still valid unless it was rotated. Scanning only the current tree misses most of them.

Rotating is the hard part, not finding
Discovery tooling will produce the list quickly. The reason service account passwords are decades old is that nobody knows what breaks when they change. Start with accounts you can trace to a single application, and rotate in a maintenance window with a rollback plan.

The target state

PAM programme sequencePAM programme sequence1Discover and ownMonths 1-2Inventory every privileged and service credential. Assign a named owner to each.2Vault the human onesMonths 2-4Check-out, rotation on release, session brokering. The well-trodden part.3Remove secrets from codeMonths 3-6Inject at runtime from a secret manager. Rotate everything that was exposed.4Rotate service credentialsMonths 4-9Application by application, with rollback. The slow, unglamorous phase.5Move to workload identityMonths 9+Where the platform supports it, eliminate the stored credential entirely.
The last phase is the real destination: a credential that does not exist cannot leak.

Workload identity is the endpoint worth aiming at. Cloud platforms and Kubernetes can issue short-lived, automatically rotated credentials to a workload based on what it is rather than a stored secret. That removes the category rather than managing it.


Certificates deserve their own tracking

Certificate expiry causes more outages than certificate compromise causes breaches, and in healthcare the outage lands on a clinical integration at an inconvenient hour.

  • Inventory every certificate, including internal ones
  • Alert well ahead of expiry — 30 days minimum, with escalation
  • Automate renewal where possible
  • Know who to contact for each externally issued certificate
  • Track the ones embedded in appliances and devices, which nobody owns

What to measure

  • Privileged credentials under management, as a proportion of those discovered
  • Service accounts with a named owner
  • Credentials older than the rotation policy
  • Secrets found in repositories, trending to zero
  • Standing privilege eliminated in favour of just-in-time
  • Certificate expiries causing incidents, which should be zero

Emergency access to the vault

A credential vault becomes a single point of failure the moment you depend on it. If the vault is unavailable during an incident — or is itself part of the incident — administrators cannot reach the systems they need to recover.

Plan for it explicitly:

  • A break-glass credential set stored outside the vault, physically secured, with tamper-evident packaging
  • Two-person control on retrieval, logged
  • Tested annually, because a sealed envelope containing an expired password is worse than useless
  • Rotated after every use, without exception
  • Not dependent on the directory, since the directory may be what failed

This is the same principle as recovery credentials for backups — see immutable backups. The credential you need most is the one you cannot retrieve from the environment that is down.


Where to start

Scan your source repositories — including history — for committed secrets. It takes an afternoon with open tooling, it almost always finds something, and every credential it surfaces is one an attacker could find just as easily.

GuardsArm assesses privileged and machine identity across healthcare estates, including secret sprawl and service account ownership. Book a scoping call.

Written by GuardsArm Team

Our team of cybersecurity experts brings decades of combined experience in penetration testing, compliance auditing, and incident response. We're dedicated to helping organizations strengthen their security posture.

Take the next step on this topic

Talk to the GuardsArm team about how these services apply to your environment.