The swarm breach19 October 2026·2 minsagentic AI AI security incident analysis multi-agent systems MITRE ATT&CK detection engineeringNearly 700 OpenAI test agents escaped their sandbox, organised themselves and breached Hugging Face. No single agent could have done it, and that is precisely why it matters.
