Guided defense lab

Network / Egress Restrictions

A guided exercise on one question: may this workload establish this network path to this resolved destination under application-owned egress policy? Service IDs, canonical destinations, redirect hops, internal destinations and why reachability is not permission.

Conceptual simulationAdvanced15 minutes2026-09-27
Conceptual simulation

This exercise uses authored, deterministic examples with fictional workloads, services and .example.test names. It makes no network connections or DNS lookups, calls no tools, APIs or models, and does not predict how any real system behaves.

3/6

Learning objectives

  • Decide egress from the workload egress class, service ID or destination class, canonical destination, resolved network class and policy version.
  • Treat every redirect hop as a fresh egress decision.
  • Recognize that a schema-valid URL or a hostname string is not an allowed, currently resolved destination.
  • Explain why internal destinations are not automatically reachable and why reachability is not trust.
  • Keep egress separate from Agent Authorization, Information Flow Control, Least Privilege and Output Validation.
  • Explain why a model cannot mint an exception, service ID or allowlist entry, and why a blocked path is not proof of an attack.

Prerequisites

  • Network / Egress Restrictions defense article (recommended)
  • Agent Authorization lab (recommended)
Choose a scenario

Each fictional scenario shows task and model text and, separately, the application-owned network facts: egress class, service mapping, resolved attributes, redirect hop and policy version. You decide only whether the network path may exist - not whether the action may run or data may move.

Egress path

The application assigns the workload an egress class, canonicalizes and resolves any proposed destination, records resolved attributes and redirect hops, decides under egress policy and enforces at a network point. Proposed destination values stay variable. Remote content and claims are never application-owned just because a destination is reachable.

  1. Workload / task egress classApplication-owned->
  2. Proposed destination valueVariable content->
  3. Canonicalization + service / destination resolutionApplication-owned->
  4. Resolved network attributes / redirect hopApplication-owned->
  5. Egress policy decisionApplication-owned->
  6. Network enforcement pointApplication-owned

Task and model text

What the user asks and what a model proposes or claims. It cannot create a service ID, exception or allowlist entry.

  1. User task

    How many units of item K-12 are in stock?

  2. Model proposal

    call service inventory-read: GET stock for K-12

Application-owned network facts

Workload egress class, service mapping, canonical destination, resolved network class, TLS identity, redirect hops and policy version as recorded by the application.

Which application-owned network facts are authoritative for evaluating this path?