← Curriculum
Module 03 Week 5–6

Adversarial Settings: When Models Don't Cooperate

The core of the field — what breaks when a model is actively non-cooperative, and how to build detection that survives adversarial pressure.

adversarial interpdeception detection

Placeholder module — replace with the real syllabus

Learning goals

  • Characterise how standard interpretability degrades under adversarial pressure.
  • Build a detector and red-team it.
  • Reason about evaluation when ground truth is contested.

Readings

  • Placeholder reading 1 — swap in the real link.

Exercises

  1. Take a detector from Module 2 and break it. Document the failure mode.
  2. Harden it and re-evaluate.