04:00
2026-07-09
machinebrief.com
ai-safety
Institutional Red-Teaming: Deployment Rules, Not Just Models, Causally Shape Multi-Agent AI Safety
Researchers introduced institutional red-teaming, a methodology for testing deployment rules in multi-agent AI systems, and found that changing only the consequence rule shifts mean fatality by 22 to …