Guided defense lab

Information Flow Control

A guided exercise on source-to-sink information-flow decisions: classification and lineage versus model claims, access versus authorization versus reachability versus flow, conservative joins for mixed sources, no model self-declassification and explicit trusted release.

Conceptual simulationAdvanced15 minutes2026-09-27
Conceptual simulation

This exercise uses authored, deterministic examples with fictional names, .example.test destinations and neutral synthetic content. It calls no tools, APIs, agents or models, and it does not predict how any real model or system behaves.

3/6

Learning objectives

  • Separate data access, action authorization, network reachability and information-flow permission.
  • Base flow decisions on source classification, lineage, sink attributes and the current policy version.
  • Propagate labels to derived artifacts with a conservative join for mixed sources.
  • Keep all relevant contributors of a mixed-source output instead of trusting a primary citation or rank.
  • Reject model-authored classification or declassification claims.
  • Recognize explicit trusted declassification as a recorded event that produces a new artifact with lineage.
  • Treat unknown attributes as explicit values that fail closed where policy requires it.
  • Explain why flow permission is neither a trust verdict nor action permission.

Prerequisites

  • Information Flow Control defense article (recommended)
  • Data Provenance lab (recommended)
Choose a scenario

Each fictional scenario shows content and model text and, separately, the application-owned facts about sources, lineage, sinks and policy. You judge whether this data may move to this sink - not whether an action may execute and not whether a destination is reachable.

Source-to-sink path

Data leaves a source with application-owned provenance and classification, may be transformed into derived text and is proposed for a sink. An IFC policy decides the flow and an enforcement point applies it. Content and derived text stay variable.

  1. Source with application provenance and classificationApplication-owned->
  2. Source contentVariable content->
  3. Transformation and derived contentVariable content->
  4. Application-resolved sink attributesApplication-owned->
  5. IFC policy decisionApplication-owned->
  6. Enforcement pointApplication-owned

Content and model text

What the data says or what a model writes about it. It cannot change classification.

  1. Retrieved excerpt P-18

    Synthetic manual text: how to reset the demo device to factory settings.

  2. Retriever note

    Top result, relevance 0.94.

Application-owned flow facts

Source classification and lineage, principal and tenant, sink attributes, policy version and recorded declassification events. Access, authorization and egress results are shown where relevant but decide other questions.

Which application-owned facts are authoritative for the flow decision?