JailbreakBench, HarmBench, StrongREJECT
Model-layer robustness, reproducibly, plus the over-reporting counterweight.
Not written yet
This module is scoped and scheduled but its content has not been written. When it ships it will cover: Model-layer robustness, reproducibly, plus the over-reporting counterweight.
Every module in the curriculum is listed from day one so the shape of the whole course is visible. An empty page is more honest than a hidden one.