Summary
This publication explains the concept of the Agentic AI harness (referred to hereafter as the harness), the software layer that enables a large language model (LLM) to interact with data, tools and systems to perform agentic tasks. It examines how harness design influences the security, governance, reliability and operational risks of agentic AI systems. ASD's Careful adoption of agentic AI services provides broader guidance on agentic AI security considerations, including expanded attack surfaces, system complexity and the evolving security landscape. It also outlines the principal risk categories of privilege, design and configuration, behavioural, structural and accountability risks, and provides best-practice guidance for designing, developing, deploying and operating secure agentic AI systems. The harness publication complements that guidance by focusing on the harness as the component organisations can most directly govern, secure and control. This publication is intended for organisations adopting or considering agentic AI.