What happened
Australia's Digital Transformation Agency (DTA) published a new 'Agentic AI Addendum' — policy guidance specifically for agentic AI systems deployed in government. Published in late June 2026, the playbook defines three oversight models: human in the loop, human on the loop, and human out of the loop, with a strong preference for continuous monitoring. Key requirements include: mandatory accountability mapping (identifying who is responsible for each agent's actions); auditable reasoning logs; memory governance; and built-in kill switches treated as design requirements, not afterthoughts. The guidance covers the full agent lifecycle from design to decommissioning. Reported on 25 June 2026 via the Public Sector Marketing Institute's AI SIX Podcast (Episode 423), which described it as sourced from the DTA.
Why it matters
This is among the first government-issued technical governance frameworks specifically addressing autonomous AI agents — a step beyond chatbot governance. It establishes binding expectations for Australian federal agencies procuring or deploying agentic AI, and is likely to influence APAC peer governments and enterprise AI governance frameworks globally. The requirement that 'the ability to stop an agent is treated as a design requirement' sets a concrete safety baseline. The framework addresses memory poisoning, tool-use controls, and human accountability chains — issues with no prior regulatory guidance in most jurisdictions.
Action needed
Australian government agencies: review the DTA Agentic AI Addendum and map all existing and planned agentic AI deployments against its three oversight model classifications; ensure kill-switch mechanisms and accountability chains are documented before deployment. Private sector: use the DTA framework as a governance benchmark for enterprise agentic AI deployments, especially where government contracts are involved.