Zero-Trust Architecture

Never trust, always verify: identity-centric access, least privilege, micro-segmentation and continuous verification, based on NIST SP 800-207, applied to users, devices and workloads.

Intent

Zero trust removes implicit trust based on network location. Every access request (user, device or workload) is authenticated, authorized and encrypted, evaluated against policy using identity and context, and granted with least privilege. NIST SP 800-207 (2020) defines the reference architecture; Google's BeyondCorp is a well-known implementation.

Core components (NIST SP 800-207)

Component Role
Policy Engine (PE) Decides to grant/deny based on policy and signals
Policy Administrator (PA) Establishes/terminates the session based on PE decision
Policy Enforcement Point (PEP) Gateway/proxy that enforces the decision in the data path
Signals Identity, device posture, threat intelligence, behaviour, data classification
flowchart LR
  SUBJ[User / device / workload] --> PEP[Policy enforcement point<br/>identity-aware proxy, mesh, API gateway]
  PEP <--> PDP[Policy decision<br/>engine + administrator]
  PDP --> SIG[Signals: IdP, device posture,<br/>risk, data classification]
  PEP --> RES[Resource: app, API, data]

Applying it

  • Users: SSO with phishing-resistant MFA (passkeys/FIDO2), conditional access by device health and risk, identity-aware proxies instead of VPN.
  • Workloads: strong workload identity (SPIFFE/SPIRE, cloud workload identity), mTLS everywhere, authorization policies per service.
  • Network: micro-segmentation; default-deny network policies; private endpoints.
  • Data: classification-driven access, encryption, DLP.
  • Continuous verification: short-lived credentials, session re-evaluation, logging every decision.

See the security view in Zero Trust Security.

Pros and cons

Pros Cons
Limits lateral movement after a breach Significant program, not a product
Works for remote users and multi-cloud Requires mature identity and device management
Fine-grained, auditable access Policy complexity; risk of blocking legitimate work

Interview questions

How would you start a zero-trust program?

Inventory users, devices, applications and data flows; consolidate identity with strong MFA; protect the most critical applications with identity-aware access first; add workload identity and mTLS for service traffic; then micro-segment and iterate, measuring coverage.

Does a service mesh give you zero trust?

It provides important pieces for workloads (identity, mTLS, service authorization), but zero trust also covers users, devices, data and continuous policy evaluation.