The deployment of autonomous AI agents in enterprise software systems has outpaced the governance infrastructure required to make such deployments safe. This paper presents COA-MAS (Cognitive Organization Architecture for Multi-Agent Systems), a governance framework grounded in organizational theory, institutional design, and normative multi-agent systems research. The framework addresses the Distributed Cognitive Chaos (DCC) failure mode — the structural consequence of deploying agents without formal authority hierarchies, authorization contracts, or enforcement boundaries — through three integrated components: 1- A four-layer organizational architecture derived from Simon's bounded rationality and hierarchical decomposition principles 2- The Action Claim ontology, a formal authorization object whose design reflects Ostrom's institutional analysis framework 3- The AASG (Autonomous Agent Security Gateway), a protocol-agnostic enforcement proxy operating at the boundary between agent coordination (A2A) and tool execution (MCP) The framework introduces the Justification Gap metric — the first computable measure of semantic proportionality between a delegated mandate and a proposed action — and the Intentionality Termination Point concept as a response to the infinite regress problem in language-based governance. The problems addressed are independently recognized by the IETF Internet-Draft on AI Agent Authentication and Authorization (draft-klrc-aiagent-auth-00, March 2026) and by enterprise security practitioners. COA-MAS implements and extends the AIMS model with a semantic authorization layer absent from existing standards.
Building similarity graph...
Analyzing shared references across papers
Loading...
Rudson Kiyoshi Souza Carvalho
Building similarity graph...
Analyzing shared references across papers
Loading...
Rudson Kiyoshi Souza Carvalho (Mon,) studied this question.
www.synapsesocial.com/papers/69ba42fb4e9516ffd37a3cd9 — DOI: https://doi.org/10.5281/zenodo.19057202