New The 2026 Continuous Validation Methodology Paper is now available. Read the paper →

Enterprise Security Architecture.

Reference architecture patterns for security at enterprise scale, paired with the cross-industry view of which platform choices the field is moving toward.

Zero-trust network access (ZTNA)

  • Pattern. No implicit trust based on network position. Every request authenticated + authorized at the resource. VPN replaced by identity-aware proxy or per-app tunnels.
  • Common failure. Legacy services bolted in via "trusted network" exception. Each exception erodes the model. Sunset list with deadlines, not allowlist with no review.
  • Vendors. Cloudflare Access, Zscaler Private Access, Tailscale, Google BeyondCorp, Microsoft Entra Private Access. Pick by identity-provider fit + protocol coverage (HTTP-only vs full L4).

Identity-aware proxy

  • Pattern. Single ingress point per application; proxy authenticates the user against IdP, attaches identity headers, forwards to backend. Backend trusts only the proxy.
  • Common failure. Backend reachable bypassing proxy (private IP exposed, container port-forward, accidental public LB). Audit by trying to hit backend directly from peer network.
  • Hardening. mTLS between proxy and backend; backend rejects any cert except proxy's. Signed identity headers (JWT) so backend can verify they came from the proxy.

Secrets-manager-backed credential flow

  • Pattern. Apps never hold long-lived secrets. App identity (workload identity, IAM role, Pod SA) → secrets manager → short-lived credential.
  • Vendors. HashiCorp Vault, AWS Secrets Manager, Azure Key Vault, GCP Secret Manager, Doppler, 1Password Connect.
  • Common failure. Bootstrap problem solved by hardcoding the bootstrap token in the AMI/image. Audit by grepping images for known token formats.
  • Rotation. Automated, scheduled, observable. Vault dynamic secrets are the canonical example — DB credentials valid for 1 hour, regenerated on demand.

Secure software delivery pipeline

  • Pattern. Code → CI → signed artifact → registry → policy gate → deploy. Every transition signed and logged.
  • Signing. Sigstore (cosign) for containers, in-toto attestations for build provenance, SLSA framework for maturity ranking.
  • Policy gates. OPA/Gatekeeper or Kyverno in K8s; admission controllers reject unsigned or non-policy-compliant deployments.
  • Common failure. CI runner has wide cloud permissions (deploys with admin role). Pipeline compromise → cloud compromise. Defense: per-job federated identity, scoped to the deploy target.

SOC integration patterns

  • Detection-as-code. Detection rules versioned in git, tested in CI, deployed via pipeline. Sigma format → SIEM-specific compilation.
  • SOAR runbooks. Each alert maps to a runbook with explicit decision points and automatable steps. Manual triage as fallback, not default.
  • Threat intel ingestion. STIX/TAXII feeds → SIEM watchlists. Quality > volume; one curated paid feed beats five free firehoses.

Regulated-industry variants

  • Finance. Heavy emphasis on segregation of duties; change management with multi-approver gates; tamper-evident logging (often append-only S3 + Object Lock).
  • Healthcare. HIPAA-driven PHI segregation; minimum-necessary access; BAA chain with every vendor that touches PHI.
  • Government. FedRAMP / IL4–IL6 deployment patterns; FIPS-validated crypto; supply-chain attestation requirements.
  • Critical infrastructure. NERC CIP / NIS2 — physical security, IT/OT segmentation, mandatory incident reporting timelines.

Technology trends (current)

  • Toward. Workload identity federation (kills static cloud creds in CI); eBPF-based runtime security; supply-chain SBOM + signing; SaaS SIEM with managed detections.
  • Away. Long-lived static API keys in CI; perimeter-only firewalls without internal segmentation; on-prem self-built SIEM without managed detection team; Java-native serialization in new code.
Rule of thumbArchitecture review evaluates the system at its weakest path, not its strongest. If 90% of access uses ZTNA but the legacy VPN still exists for ops, the security posture is determined by the VPN, not the ZTNA. Find the bypass before assuming the pattern works.

From reference to evidence

Run this against your own environment.