Tag
This paper introduces institutional red-teaming, an evaluation methodology for testing deployment rules in multi-agent AI systems, showing that deployment rules causally affect safety outcomes and that identity salience can drive targeted elimination in agent populations.