Agent Shared Security and Safety Responsibility Model (Agent 3SRM)

As AI agents transition from experimental prototypes to production enterprise systems, the question of who is responsible for what becomes critical. The CSA AI Controls Matrix (AICM) v1.1 establishes a shared responsibility framework for AI systems, distributing 247 control objectives across five supply chain roles: Cloud Service Provider (CSP), Model Provider (MP), Orchestrated Service Provider (OSP), Application Provider (AP), and AI Customer (AIC). However, the AICM was designed for the general AI/ML service delivery model. Agentic AI systems, with their autonomous operation, sub-agent delegation, persistent memory, and cross-organizational tool use, introduce responsibility and accountability attribution challenges that extend beyond the AICM’s current scope. This paper presents the Agent Shared Security and Safety Responsibility Model (Agent SSSRM/3SRM), introduces the Agent Deployment Model and extends the AICM’s five-role value chain framework for ML/LLM and mapping it to the ten-layer AI Agent Reference Architecture defined in the companion paper. The Agent 3SRM maps the AICM’s control ownership assignments to each of the 10 layers Agentic Reference Architecture, defined in the companion paper, identifies agentic-specific responsibility gaps. Finally the paper introduces the concept of the Agent Owner as the entity that bears ultimate, non-delegable accountability for all agent actions and defines accountability chains for sub-agent delegation. The result is a practical framework that bridges the AICM’s control-level responsibility model with the architectural reality of deployed agent systems.