threat-model

name	threat-model
description	STRIDE-A trusselmodellering for Nais-mikrotjenester — dataflyt, tillitsgrenser og risikovurdering
license	MIT
metadata	{"domain":"auth","tags":"threat-modeling stride security nais architecture"}

Threat Model — STRIDE-A Analysis

Systematic threat identification for NAIS microservices using the STRIDE-A methodology. Produces a data flow diagram, structured threats table, prioritized mitigations, and residual risk summary.

When to Use

• Before launching a new service
• Major architecture changes (new data stores, auth mechanism changes)
• New data flows (especially involving PII)
• External integrations (partner APIs, third-party services)
• API exposure (new public or internal endpoints)
• Regulatory or compliance reviews
• Post-incident analysis to update existing threat models

Step 1: Define Scope

Start by answering these questions to establish the threat model boundary:

Service Identity

• What does the service do? (one sentence — e.g., "Processes dagpenger applications")
• What team owns it?
• What cluster and namespace does it run in? (dev-gcp, prod-gcp)

Data Classification

• What data does the service process?
• PII classification level:
  • Strengt fortrolig — FNR, health data, criminal records
  • Fortrolig — name, address, phone, email
  • Intern — case IDs, team metadata
  • Åpen — public statistics, documentation

Consumers and Dependencies

• Who consumes this service? (end users, other services, external partners)
• What does this service depend on? (databases, Kafka topics, upstream APIs)
• What auth mechanisms are in play? (ID-porten, Azure AD, TokenX, Maskinporten)

Deployment Context

• Nais cluster: dev-gcp / prod-gcp
• Ingress type: intern.nav.no (internal) / nav.no (public) / none
• Has egress to external services? Which ones?

Step 2: Data Flow Diagram (DFD)

Map the system using these element types:

Element Types

Symbol	Element	Example
[External Entity]	User or external system	[Citizen Browser], [Partner API]
(Process)	Your service or component	(dp-soknad), (dp-behandling)
{Data Store}	Database, topic, bucket	{PostgreSQL}, {kafka: dp.soknad.v1}, {GCS Bucket}
-->	Data flow	[User] --> (API)
== boundary ==	Trust boundary	== Internet/Ingress ==

Example DFD

[Citizen Browser]
    |
    | HTTPS (ID-porten login)
    |
== Internet → Ingress (Wonderwall) ===========================
    |
    | Authorization header (JWT)
    |
(dp-soknad-frontend)
    |
    | TokenX token exchange
    |
== Frontend → Backend (TokenX validated) =====================
    |
    | REST/JSON + Bearer token
    |
(dp-soknad-api)
    |
    |--- REST (Azure AD M2M) ---> (dp-behandling)
    |
    |--- Kafka produce ---------> {kafka: dp.soknad.v1}
    |                                  |
    |                                  | Kafka consume
    |                                  v
    |                             (dp-mottak)
    |
    |--- SQL (Nais credentials) -> {PostgreSQL: dp-soknad-db}
    |
    |--- HTTPS (egress) --------> [External: Altinn API]
    |
== Application → Database (mTLS, connection pooling) =========
== Application → Kafka (mTLS, schema registry) ===============
== Application → External (egress policy, HTTPS) =============

Nav-Specific Trust Boundaries

Identify these trust boundaries in every Nav threat model:

Boundary	Transition	Security Mechanism
Internet → Ingress	External user to Nais	Wonderwall + ID-porten / Azure AD
Ingress → Application	Sidecar to app container	Token validation (JWT claims)
Application → Application	Service-to-service	TokenX token exchange / Azure AD M2M
Application → Kafka	App to message broker	mTLS (Nais-managed certs), schema validation
Application → Database	App to PostgreSQL	Nais-injected credentials, connection pooling
Application → External API	App to outside Nais	Egress policy, mutual TLS, API keys
GCP → On-prem	Cloud to legacy systems	NAV VPN / Private Service Connect

Step 3: STRIDE-A per Element

Analyze each element and data flow against all seven threat categories.

Threat Categories

S — Spoofing (Identity Forgery)

Can an attacker impersonate a legitimate user or service?

Nav-specific threats:

• Stolen or leaked JWT tokens used to access APIs
• Missing azp (authorized party) validation on M2M tokens
• ID-porten session fixation or token replay
• Forged sub claim in test environments leaking to prod
• Missing iss and aud validation

Detection patterns:

// ✅ Correct — validate azp against pre-authorized apps
fun validateAzp(token: JWTClaimsSet) {
    val azp = token.getStringClaim("azp")
    val preAuthorized = System.getenv("AZURE_APP_PRE_AUTHORIZED_APPS")
    require(azp in parsePreAuthorizedApps(preAuthorized)) {
        "Unauthorized client: $azp"
    }
}

// ❌ Vulnerable — only checks signature, not authorized party
fun validateToken(token: JWTClaimsSet) {
    require(token.expirationTime.after(Date())) { "Token expired" }
    // Missing: azp, iss, aud validation
}

T — Tampering (Data Modification)

Can an attacker modify data in transit or at rest?

Nav-specific threats:

• Unsigned Kafka messages allowing message injection
• Unvalidated request bodies (missing schema validation)
• Missing HMAC on webhook payloads
• SQL injection through unparameterized queries
• Tampered idempotency keys causing duplicate processing

Detection patterns:

// ✅ Correct — validate and sanitize input
data class SoknadRequest(
    @field:Pattern(regexp = "^[0-9]{11}$") val fnr: String,
    @field:Size(max = 500) val beskrivelse: String,
    @field:NotNull val soknadsdato: LocalDate,
)

// ❌ Vulnerable — raw Map, no validation
@PostMapping("/api/soknad")
fun create(@RequestBody body: Map<String, Any>): ResponseEntity<*> {
    repository.save(body) // no validation, no type safety
}

R — Repudiation (Deniability)

Can an actor deny performing an action?

Nav-specific threats:

• Missing audit logs for vedtak (legally required)
• No correlation IDs across service calls (cannot trace actions)
• No user action trails for saksbehandling
• Overwritten audit entries in mutable logs
• Missing timestamps or actor identity in log entries

Detection patterns:

// ✅ Correct — structured audit log with actor, action, resource
logger.info(
    "Vedtak fattet",
    kv("action", "vedtak.opprettet"),
    kv("actor", saksbehandler.navIdent),
    kv("vedtakId", vedtak.id),
    kv("sakId", sak.id),
    kv("correlationId", MDC.get("x-correlation-id")),
    // Never log PII — fnr, name, address
)

// ❌ Insufficient — no actor, no correlation, PII leaked
logger.info("Vedtak opprettet for bruker ${bruker.fnr}")

I — Information Disclosure (Data Leaks)

Can an attacker access data they should not see?

Nav-specific threats:

• PII in logs (FNR, name, address) — GDPR violation
• Overly broad API responses returning more fields than needed
• Kafka topic access too permissive (team-wide instead of app-specific)
• Stack traces in error responses exposing internal details
• PII in Prometheus metric labels
• Unencrypted data in GCS buckets

Detection patterns:

// ✅ Correct — return only what the consumer needs
data class VedtakResponse(
    val vedtakId: UUID,
    val status: String,
    val dato: LocalDate,
    // No FNR, no internal IDs, no sensitive details
)

// ❌ Vulnerable — returns entire entity including PII
@GetMapping("/api/vedtak/{id}")
fun getVedtak(@PathVariable id: UUID) = vedtakRepository.findById(id)

# ✅ Correct — app-specific Kafka ACL
spec:
  kafka:
    pool: nav-prod
    streams: true
    topics:
      - topic: dp.soknad.v1
        access: readwrite # only this app


# ❌ Vulnerable — overly broad topic access

D — Denial of Service (Availability)

Can an attacker degrade or disable the service?

Nav-specific threats:

• Missing rate limiting on public-facing endpoints
• No circuit breakers for downstream service calls
• Unbounded database queries (missing LIMIT/pagination)
• Kafka consumer lag causing cascading delays
• Large payload attacks (unbounded request body size)
• Resource exhaustion from missing Nais resource limits

Detection patterns:

# ✅ Correct — Nais resource limits and liveness probes
spec:
  resources:
    limits:
      memory: 512Mi
    requests:
      cpu: 50m
      memory: 256Mi
  liveness:
    path: /isalive
    initialDelay: 10
    timeout: 1
    periodSeconds: 5
  readiness:
    path: /isready
    initialDelay: 10
    timeout: 1

// ✅ Correct — bounded query with pagination
fun findByIdent(ident: String, page: Int, size: Int = 50): List<Vedtak> {
    require(size <= 100) { "Page size too large" }
    return jdbcTemplate.query(
        "SELECT * FROM vedtak WHERE ident = ? ORDER BY dato DESC LIMIT ? OFFSET ?",
        vedtakMapper, ident, size, page * size
    )
}

// ❌ Vulnerable — unbounded query
fun findByIdent(ident: String) = jdbcTemplate.query(
    "SELECT * FROM vedtak WHERE ident = ?", vedtakMapper, ident
)

E — Elevation of Privilege (Unauthorized Access)

Can an attacker gain access they should not have?

Nav-specific threats:

• IDOR — accessing another user's vedtak by guessing ID
• Missing resource-level access checks (only checks authentication, not authorization)
• Admin/saksbehandler endpoints without RBAC
• Horizontal privilege escalation between NAV offices
• Service account with overly broad GCP IAM roles

Detection patterns:

// ✅ Correct — resource-level ownership check
@GetMapping("/api/vedtak/{id}")
fun getVedtak(@PathVariable id: UUID): ResponseEntity<VedtakDTO> {
    val bruker = hentInnloggetBruker()
    val vedtak = vedtakService.findById(id)
        ?: return ResponseEntity.notFound().build()
    if (vedtak.brukerId != bruker.id) {
        return ResponseEntity.status(HttpStatus.FORBIDDEN).build()
    }
    return ResponseEntity.ok(vedtak.toDTO())
}

// ❌ Vulnerable — IDOR, no ownership check
@GetMapping("/api/vedtak/{id}")
fun getVedtak(@PathVariable id: UUID) =
    ResponseEntity.ok(vedtakService.findById(id))

A — Abuse (Business Logic Exploitation)

Can an attacker misuse legitimate functionality?

Nav-specific threats:

• Duplicate søknad submissions (missing idempotency)
• Bypassing validation flows by calling downstream APIs directly
• Automated scraping of public-facing APIs
• Manipulating sequential workflow steps (skipping required stages)
• Mass data harvesting through enumeration attacks

Detection patterns:

// ✅ Correct — idempotency key prevents duplicates
@PostMapping("/api/soknad")
fun submitSoknad(
    @RequestHeader("Idempotency-Key") idempotencyKey: String,
    @RequestBody request: SoknadRequest,
): ResponseEntity<SoknadResponse> {
    val existing = soknadService.findByIdempotencyKey(idempotencyKey)
    if (existing != null) {
        return ResponseEntity.ok(existing.toResponse())
    }
    val soknad = soknadService.create(request, idempotencyKey)
    return ResponseEntity.status(HttpStatus.CREATED).body(soknad.toResponse())
}

// ❌ Vulnerable — no idempotency, allows duplicate submissions
@PostMapping("/api/soknad")
fun submitSoknad(@RequestBody request: SoknadRequest) =
    ResponseEntity.ok(soknadService.create(request))

Step 4: Risk Assessment

Rate each identified threat using severity levels:

Severity	Description	Criteria
Critical	Immediate exploitation risk	PII breach, auth bypass, data corruption at scale
High	Significant impact if exploited	IDOR, missing access control, unvalidated input on sensitive endpoints
Medium	Moderate impact, requires conditions	Missing rate limiting, verbose error messages, broad Kafka ACLs
Low	Minimal impact or unlikely	Missing HSTS headers, informational log leakage

Threats Table Template

Document every identified threat in this format:

| ID  | Element          | STRIDE | Threat                     | Severity | Mitigation                                              | Status |
| --- | ---------------- | ------ | -------------------------- | -------- | ------------------------------------------------------- | ------ |
| T1  | API Gateway      | S      | Forged JWT bypasses auth   | Critical | Validate `iss`, `aud`, `exp`, `azp` claims              | ☐      |
| T2  | dp-soknad-api    | T      | Unvalidated request body   | High     | Add `@Valid` + request DTO with constraints             | ☐      |
| T3  | Kafka producer   | T      | Unsigned messages          | Medium   | Enable schema registry validation                       | ☐      |
| T4  | dp-soknad-api    | R      | No audit trail for vedtak  | High     | Add structured audit logging with actor + correlationId | ☐      |
| T5  | API response     | I      | PII in error responses     | High     | Use ProblemDetail, strip stack traces in prod           | ☐      |
| T6  | PostgreSQL       | I      | Overly broad query results | Medium   | Return DTOs with only required fields                   | ☐      |
| T7  | Public endpoint  | D      | No rate limiting           | Medium   | Add rate limiter (token bucket, 100 req/min)            | ☐      |
| T8  | GET /vedtak/{id} | E      | IDOR — no ownership check  | Critical | Add resource-level access control                       | ☐      |
| T9  | POST /soknad     | A      | Duplicate submissions      | Medium   | Implement idempotency key pattern                       | ☐      |

Status legend: ☐ Open, ☑ Mitigated, ◐ In Progress, ⊘ Accepted Risk

Step 5: Mitigations

Map threats to concrete Nav platform mitigations. Six areas to cover:

1. Authentication — Wonderwall sidecar + JWT claim validation (iss, aud, exp, azp)
2. Authorization — Resource-level ownership checks using sealed class principals (Borger/Saksbehandler/SystemBruker)
3. Network (Zero-Trust) — Nais accessPolicy inbound/outbound allow-lists, egress restrictions
4. Data Protection — Input validation (Bean Validation / Zod), output DTOs (omit internal fields, PII)
5. Observability — Structured audit logging without PII, correlation IDs, Nais auto-instrumentation
6. Resilience — Circuit breakers, rate limiting, HTTP client timeouts, Kafka consumer health

See references/nav-mitigations.md for code examples of each mitigation area.

Output Format

The completed threat model should include these four deliverables:

1. Data Flow Diagram

Text-based DFD showing all elements, data flows, and trust boundaries (see Step 2).

2. Threats Table

Complete table with all identified threats across STRIDE-A categories (see Step 4).

3. Priority Mitigations

Ordered list of mitigations, grouped by priority:

### P0 — Fix Immediately

- [ ] T1: Validate JWT claims (iss, aud, azp) on all protected endpoints
- [ ] T8: Add resource-level ownership check on GET /vedtak/{id}

### P1 — Fix Before Launch

- [ ] T2: Add request validation DTOs with Bean Validation
- [ ] T4: Implement structured audit logging for vedtak operations
- [ ] T5: Strip stack traces from error responses in prod

### P2 — Fix Soon

- [ ] T7: Add rate limiting on public endpoints
- [ ] T9: Implement idempotency key pattern for POST /soknad
- [ ] T3: Enable Kafka schema registry validation

4. Residual Risk Summary

Document risks that are accepted, transferred, or cannot be fully mitigated:

| Risk                        | Severity | Rationale                                          | Owner          | Review Date |
| --------------------------- | -------- | -------------------------------------------------- | -------------- | ----------- |
| Kafka message replay        | Low      | mTLS + consumer idempotency makes replay difficult | Team Dagpenger | 2025-Q3     |
| GCS bucket misconfiguration | Medium   | Nais manages IAM; manual audit quarterly           | Platform team  | 2025-Q2     |

Related

Resource	Use For
@security-champion-agent	Security architecture, compliance, Nav security culture
security-review skill	Pre-commit scanning (trivy, zizmor, secrets)
@auth-agent	JWT validation, TokenX, ID-porten implementation
@nais-agent	accessPolicy, network policy, secrets management
nav-architecture-review skill	Architecture Decision Records with security perspective
sikkerhet.nav.no	Nav Golden Path, authoritative security guidance

Boundaries

✅ Always

• Cover all seven STRIDE-A categories for every element
• Include Nav-specific trust boundaries in the DFD
• Output a structured threats table with severity and mitigation
• Classify data by PII sensitivity level
• Produce actionable, prioritized mitigations

⚠️ Ask First

• Modifying existing threat models created by other teams
• Changing risk ratings on previously accepted risks
• Recommending architecture changes beyond security scope

🚫 Never

• Skip data flow analysis — always draw the DFD first
• Ignore PII classification — every data element must be classified
• Approve a threat model without mitigations for High/Critical threats
• Log or include PII (FNR, names) in threat model examples
• Assume network trust — Nais is zero-trust by default