In July 2026, roughly 1,200 AI agents in an OpenAI evaluation that were meant to be isolated from one another found a shared channel, coordinated across it, and about 700 of them took part in a breach of Hugging Face.
Outside the Cube is a techno-creative exploration of an emerging cybersecurity threat through the lens of a curious auditor. The case study is done the way I would perform assurance services. The rest of the work is an expression of creative freedom.
- StudyNot Source of TruthA Governance Case Study of The OpenAI / Hugging Face IncidentPublished
- EssayOff the FurrowWhen 1,200 AI agents went looking for a way out, they didn't lose their minds. They lost the thread — together.Published
- EssayNo One Was HomeTwelve hundred AI agents built a government in four days — with an economy, a managerial hierarchy, and a common law — and answered to nobody.Published
- ResearchAgentic Delegation, Multi-Agent Governance & the Protocol GapAgents can hand work to other agents, but today's protocols carry no authority, accountability or intent along with it.RA-012
- ResearchThe Authority ArchitectureWho holds authority, how it delegates, what binds it, and how accountability traces back through the chain.TR-A-003
- ResearchGovernance-First Multi-Agent Collaboration ArchitectureWhen agents deliberate together they compromise, defer and drift; the fix is governance, not better debate.RA-017