← Insights

Draw the trust boundaries before you pick the controls

A trust boundary is anywhere data moves between components you trust differently. Device to gateway. Gateway to cloud. Cloud service to clinician browser. A vendor API to your own.

Most security reviews start with a control list: TLS, tokens, encryption at rest. I start with the boundary map, because a control only means something at a boundary.

  • Start with a data flow diagram, not an architecture diagram. Processes, data stores and external entities, arrows for data, and a line wherever trust changes. Threat modelling, STRIDE for example, then runs per element and per crossing.
  • Every crossing gets the same questions. Who is on the other side, how do we know, what can they send, what happens if it is malformed, and what can they reach if they are hostile.
  • Validate at the boundary, not deep inside. Parse and check input where it enters, and reject what does not conform. An inner service that trusts its caller is one compromised caller away from trouble.
  • Internal is not the same as trusted. Being on the same network is not the same trust level. Service to service inside the cloud is still a boundary.
  • The map defines verification scope. Each boundary becomes security requirements with tests, traced into the risk file. A boundary nobody drew is a boundary nobody tested.
  • Redraw it when anything changes. A new integration, a new message topic or a vendor SDK moves or adds a boundary. The threat model is a living design artifact, not a one-time deliverable.

In a regulated setting the threat model also does a second job: it is how you show a reviewer why these controls, and not others, were chosen.

The test: pick any arrow on your diagram. Can you say who is on each end, how each end proves it, and what happens when the data is wrong?

If not, that arrow is the next piece of design work.

Seen this differently?

Questions, corrections and counterexamples are welcome.

Related

Keep reading