Insights · Security Architecture
Zero trust is an architecture decision, not a product purchase
Zero trust lives in decisions no vendor can sell: how services authenticate, where secrets live. Retrofitting them later is a rewrite.
The premise, stated precisely
“Zero trust” has been marketed into mush, so let’s restate it in engineering terms: no request is trusted because of where it came from. Network location grants nothing — NIST’s definition in SP 800-207 is exactly this: “no implicit trust granted to assets or user accounts based solely on their physical or network location.” Every request, user to service, service to service, service to database, carries verifiable identity, and authorization is evaluated per request against policy — the UK NCSC compresses the same posture: “the network is assumed hostile, and each request is verified based on an access policy” (NCSC, “Zero trust architecture design principles”).
Stated that way, zero trust is visibly not a box. It’s a set of architecture decisions with cost and consequence, and they belong in your decision records like any other.
The decisions that actually are zero trust
On the E-Tax modernization for Rwanda Revenue Authority, these were explicit, governed decisions, several with their own ADRs:
- Service identity. How does the Registration service know a request from the Payment service is genuine? On the target platform: mutual TLS for all east-west traffic, issued and rotated automatically by the service mesh — the mesh “securely provisions strong identities to every workload with X.509 certificates” and its agents “automate key and certificate rotation at scale” (Istio documentation, security concepts). A policy decision, then a configuration one. This was ADR material, not a product sheet.
- Edge policy. Gateway security and defense-in-depth at the ingress boundary (WAF, rate limiting, authentication), governed as its own decision set (gateway strategy was literally the programme’s first ADR).
- Secrets. Where credentials, keys and certificates live, who can read them, and for how long. OWASP’s Secrets Management Cheat Sheet opens with the rule to “centralize and standardize” secret storage (OWASP); centralized management with dynamic credentials and automated rotation follows, including the uncomfortable interim questions like how the secrets store itself unseals after a restart (the production guide on this site covers the runtime side).
- Data protection. Encryption in transit and at rest is table stakes; the real decisions are retention integrity (audit trails designed tamper-evident, with 15-year retention on E-Tax) and privacy law mapped into the architecture: whose data may cross which boundary is an architecture constraint, not a compliance afterthought.
- Runtime detection. What watches workloads after they pass the pipeline. Runtime security — Falco-class syscall-level detection (Falco documentation) — closes the loop that build-time scanning leaves open.
- Identity federation. One identity fabric (OAuth2/OIDC, MFA, external federation) serving users and services, not three directories that disagree — built on the standard authorization framework (OAuth 2.0) so tokens, scopes and lifetimes are policy, not folklore.
Why retrofitting fails
Each decision above touches the write path of every service. Retrofitting service-to-service authentication onto a platform of mutual network trust means touching every deployment, every environment and every integration partner: coordinated change across the whole estate, which is to say: a rewrite wearing a security budget’s clothes. NIST’s cloud-native extension, SP 800-207A, frames the target — “remove the implicit trust in users, services, and devices based only on their network location” — a property you design in, not bolt on.
That’s why sequencing matters more than tooling. On a modernization programme, zero trust belongs in Phase 1’s foundation: the platform layer exists before the business services land on it, so services are born with mesh-issued identity rather than granted it later. For an existing platform, the next section is the honest path.
Where to start if you’re mid-platform
- Inventory trust assumptions. List what authenticates what today. Most platforms find service-to-service calls with no authentication at all; the finding itself justifies the work.
- Centralize secrets first. It’s the least invasive decision with the most immediate risk reduction, and it teaches the organizational muscle (rotation, policy, audit) the rest will need.
- Mesh one service pair, not the estate. Prove mTLS issuance, rotation and policy on a contained pair; expand by domain. This is defense-in-depth in project form.
- Write the ADR before the purchase order. The decision record forces the requirement to be stated; the product conversation gets better when it starts from requirements instead of logos (how to write one).
Discuss your security architecture
Sources
- NIST SP 800-207, “Zero Trust Architecture” (Rose, Borchert, Mitchell, Connelly; 2020)
- NIST SP 800-207A, “A Zero Trust Architecture Model for Access Control in Cloud-Native Applications in Multi-Cloud Environments” (Chandramouli & Butcher, 2023)
- UK NCSC, “Zero trust architecture design principles”
- Istio documentation — security concepts: workload identity, mutual TLS, certificate rotation
- OWASP Secrets Management Cheat Sheet
- Falco documentation — runtime threat detection
- OAuth 2.0 — the OAuth 2.0 Authorization Framework (oauth.net)