NSW launches Australian-first agentic AI guidelines for government
NSW has released new guidance for using agentic AI across the public sector, marking an Australian first. The announcement was made on 20 October 2025 at the University of Technology Sydney.
Unlike basic automation, agentic AI can plan, make decisions, and carry out tasks. It's already being tested in finance for fraud detection and in e-commerce for content and layout optimisation-now the focus turns to safe use in government.
What the guidance covers
The new guide sets out clear expectations for safe and effective adoption. It emphasises human oversight and practical controls at every step.
- Define when and how human intervention is required.
- Run regular reviews of outputs and system behaviour.
- Apply privacy and security controls from day one.
- Use checklists to assess whether an AI agent is needed at all.
- Follow structured steps before piloting and deploying.
"I believe as a government we have a responsibility to make sure [AI] benefits people and that it is used responsibly and ethically," said NSW Minister for Customer Service and Digital Government, Jihad Dib. "The agentic AI guide will help NSW Government agencies use this developing technology safely and effectively… and keep human oversight at the forefront of any decision-making technology."
Governance and assurance
All NSW Government programs using agentic AI must be evaluated against the AI Assessment Framework, which is being updated to include agent-specific guidance. High-risk projects will go to the AI Review Committee.
"The newly formed Office for AI is hitting the ground running and is putting the safe and ethical use of AI at the centre of government decision making," said Acting Head of Digital NSW, Narelle Underwood. "The new AI agent guide and the updated AI Assessment Framework… will help ensure NSW continues to deploy this technology responsibly."
What agencies should do next
- Identify candidate use cases with clear public value and measurable outcomes.
- Map human-in-the-loop checkpoints and escalation paths for sensitive decisions.
- Set monitoring cadences, logging, and audit trails before any pilot.
- Run privacy and security checks with your cyber and data teams early.
- Start small with controlled pilots, defined exit criteria, and stakeholder communication.
- Upskill teams on agent safety, evaluation, and prompt/interaction guidelines.
Where to read the guide
Access the NSW agentic AI guide here: Guide to using AI agents in NSW Government.
Related governance materials: AI Assessment Framework.
The Department of Customer Service will continue to refine this guidance as global standards progress.
Build capability
If you're planning workforce upskilling for AI projects and oversight, explore practical training options by role: AI courses by job.
Your membership also unlocks:
