Notes on multi-agent red-teaming.
A short note on emergent harm in populated agent environments — and why a single-agent safety test won't catch it.