Limites de confiance et réduction de la surface d’attaque

A trust boundary is any point where data or control crosses between zones of different trust levels. Examples include the line between the public internet and your API gateway, or between your application and a third-party service.

Every time data crosses a boundary, you must validate and authorize it. Insecure design often comes from assuming data inside a boundary is automatically safe.

Attackers exploit the assumption that internal callers are trustworthy. If a microservice trusts another service blindly, a single compromised node can pivot across your whole system.

• Treat each boundary crossing as a fresh validation point
• Never reuse trust from one layer to skip checks in another
• Document boundaries explicitly in your architecture

The attack surface is the sum of all points where an attacker can try to enter or extract data: open ports, API endpoints, input fields, file uploads, environment variables, and dependencies.

A smaller surface means fewer things to defend and fewer ways to fail.

Start by enumerating every entry point. A simple inventory helps you reason about exposure.

Dead code and forgotten endpoints are prime targets. The most effective surface reduction is deletion: remove debug routes, unused admin panels, and legacy API versions.

If you do not need it in production, it should not be reachable in production.

Apply the principle of least functionality: each component exposes only the features it truly needs. Disable directory listing, sample apps, verbose error pages, and unused protocol handlers.

• Close ports you do not use
• Disable HTTP methods you do not implement
• Strip development tooling from production images

Place databases and internal services behind network boundaries so they are not reachable from the internet. Use private subnets, security groups, and firewall rules so each tier only talks to the tier it must.

Segmentation turns a single breach into a contained incident instead of a full compromise.

When a request crosses into your service, re-validate authentication, authorization, and input shape even if an upstream layer claims to have done so.

A Data Flow Diagram (DFD) visualizes processes, data stores, external entities, and the trust boundaries between them. Drawing boundaries as dashed lines on a DFD makes it obvious where validation must happen.

DFDs feed directly into threat modeling: each boundary crossing is a candidate for STRIDE analysis.

External services, SDKs, and APIs sit on the far side of a trust boundary. Validate their responses, set timeouts, and never embed secrets that grant more access than needed.

• Treat third-party responses as untrusted input
• Use scoped, least-privilege credentials
• Fail safely when a dependency misbehaves

Attack surface grows over time as features are added. Make surface review part of design reviews and release checklists so new endpoints, ports, and dependencies are deliberately evaluated, not accidentally exposed.

You learned to identify trust boundaries, map the attack surface, and reduce it through deletion, least functionality, and network segmentation. Re-validate at every boundary, treat third parties as untrusted, and review the surface continuously as the system evolves.