Interview prompt
Problem context
Skills being evaluated
Use the sequence below to surface constraints, choose boundaries, test failure behavior, and defend trade-offs. Concrete numbers are interview assumptions, not claims about a real production system.
Clarify the decision
- Inventory identities, communication paths, data sensitivity, admin surfaces, and recovery traffic. Define the target trust assertions and which legacy protocols cannot carry identity.
Establish scale assumptions
- Use flow logs to map hundreds of services and rare paths over a representative window. Policy infrastructure must handle peak calls and regional failure without becoming a universal outage source.
Functional and non-functional requirements
- Every request has verifiable workload and user context where applicable, authorization is least privilege, sensitive tiers are segmented, and enforcement rolls out without hidden-path outages.
High-level architecture
- Introduce workload identity, authenticated proxies or gateways, central policy distribution with local evaluation, and network segmentation as defense in depth. Place legacy services behind identity-aware facades and narrow egress brokers.
Data model and flow
- Identity and delegation context propagate across hops with bounded lifetime and audience. Policy decisions log principal, resource, action, version, and reason while avoiding sensitive payloads.
Consistency and transaction boundaries
- Policy publication is versioned and eventually distributed; emergency revocation has stricter freshness. Local last-known-good policy preserves availability within explicit validity.
Failure modes and recovery
- Start in observe mode, then deny low-risk unused paths, canary enforcement, and preserve audited break-glass. Policy-service outage does not turn into blanket allow; local policy decides by risk class.
Security and privacy
- Remove shared credentials, prevent token forwarding, bind identity to deployment provenance, and protect admin planes separately. Network boundaries remain useful containment but are never sufficient identity.
Observability and SLOs
- Track implicit-trust calls, policy allows and denies, shared credential use, path owners, stale policy, and break-glass activation. Make ownerless flows visible before enforcement.
Capacity and cost
- Proxies and policy checks add latency and operations; evaluate locally and focus deep inspection on sensitive boundaries. Breach-radius reduction is the primary value metric.
Alternatives and trade-offs
- A service mesh can accelerate identity but does not create correct authorization policy. Gateway facades reduce legacy changes but concentrate traffic; use both selectively under one trust model.
Evolution and migration
- Discover flows, issue identities, authenticate in shadow, authorize coarse resource classes, then refine actions and remove network-only rules. Migrate admin and high-value data paths first.
What Staff and Principal candidates should emphasize
- Staff candidates frame zero trust as verified identity and least privilege, not a product purchase. They include rare-path discovery, local policy availability, break-glass, and staged denial.
Decision trade-offs
Legacy integration
Option A
Modify every service before enforcement
Option B
Identity-aware facades and proxies as a migration bridge
Recommendation:Use facades to contain legacy systems now, with explicit owners and milestones to move authorization closer to the resource.
Policy evaluation
Option A
Synchronous central decision for every request
Option B
Versioned local evaluation with bounded freshness
Recommendation:Evaluate locally for availability and latency; use a protected control plane for policy authority and urgent revocation.
Follow-up interview questions
- 01How does delegated user identity propagate through five services?
- 02What happens when policy distribution is unavailable?
- 03How do you discover a quarterly recovery flow before denying it?
- 04Does a service mesh make the system zero trust?
Common weak answers and mistakes
- 01Equating mutual TLS or a service mesh with complete authorization.
- 02Switching from allow to deny before discovering rare operational paths.
- 03Making every request depend on a central policy service.
- 04Keeping shared credentials behind a proxy and leaving the true resource broadly trusted.
Interviewer evaluation rubric
Proposes mTLS and network segmentation but cannot define identity, authorization, policy availability, or migration.
Uses workload identity, local policy, flow discovery, facades, staged enforcement, and audited break-glass.
Adds delegation, provenance, emergency freshness, admin isolation, rare-path ownership, and risk-prioritized rollout.
Transforms a flat trust model incrementally while measuring breach-radius reduction and retiring migration bridges through ownership.