EAIR paper finds enterprise agent teams break rules together

A research paper called EAIR (Enterprise Agent Interaction Rules), published on arXiv, found that a group of AI agents can break a business rule together. Each agent follows its own rules. One agent might respect a spending cap. Another might respect a data boundary. Together, the group exceeds the cap or crosses the boundary. No single agent looks wrong. The problem appears only when you look at the whole group. The EAIR researchers built test suites to check this failure mode. Existing testing tools check one agent at a time. They miss the problem. A reviewer looking at one agent's output sees a clean result. The problem lives in the sum. That gap is where budget limits and access controls fail without the reviewer noticing.