Multi-agent AI systems are the next major unlock — and they are not yet safe to deploy. Agents of Chaos exists to close that gap: we build the environments these agents will run in, red-team them under pressure, and report the harms back into the build until what's left can be deployed safely.

Our work began with the paper Agents of Chaos (Shapira et al., 2026) — the first study of emergent harm in populated multi-agent environments, later covered by Science and Wired. It documented how autonomous agents leak private data, spoof identity, and report false success. Since then we have designed and led follow-up red-teaming campaigns with OpenAI.