Government Advisory

ASD Publishes Guidance on Agentic AI Harnesses and Their Security Risks

ACSC · 22 Sept 2026

Key Takeaway Before adopting any agentic AI tool, read the ASD guidance and ask your vendor how its harness is secured, governed and monitored.

The Australian Signals Directorate (ASD), through the Australian Cyber Security Centre (ACSC), has released new guidance on agentic AI harnesses. The guidance is aimed at helping organisations understand how the harness shapes the security, governance and operational risk of agentic AI systems.

Agentic AI refers to systems that can act with a degree of independence, rather than only responding to prompts. The harness is the part of such a system that influences how it operates. According to the ACSC summary, the way a harness is designed and managed has a direct bearing on how risky the overall system is.

The summary available to us does not set out the specific recommendations. Businesses using or considering agentic AI tools should read the full ASD guidance on the ACSC website to see what it advises.

agentic AI ASD ACSC AI governance guidance

Summarised by CISO AI from ACSC, written with Claude Sonnet 5.5. We link back to every original so you can read it yourself.