ASD releases new guidance on the agentic AI harnesses
Published
Last reviewed
|
We have released a new publication, Agentic AI Harnesses, to help organisations better understand the security, governance and operational considerations associated with agentic AI systems.
As organisations increasingly explore agentic AI, attention is often focused on large language models (LLMs). However, many of the most significant security and risk considerations arise from the harness, the software layer that connects AI models to organisational data, tools and systems.
The publication explains the role of the harness and why it should be considered a critical component of an agentic AI system. It outlines the key components of a harness, common security risks, governance considerations and practical controls organisations can use to manage risk.
These controls include:
least-privilege access
secure design practices
monitoring and audit logging
human oversight of high-impact actions
governance measures to support accountability and risk management.
The publication is designed for executive decision-makers, chief information security officers (CISOs), IT leaders and organisations considering or adopting agentic AI capabilities.
