NSW Guide to using AI Agents in NSW Government the first AU public-sector framework built for agents

Why it matters for NETEVO

The first Australian public-sector framework written specifically for the deployment of AI agents rather than for AI in general. Issued in October 2025, the guide sets out a six-element framework — a policy position, use case identification, ownership and guardrails before launch, safe pilots, production scaling, and an understanding of the unique risks of the technology — that serves as the operational checklist for agencies considering AI agent deployment. Three features give it that standing.

An AU state-government voice naming the agentic-authority failure surface. The guide articulates, in NSW Government voice, the same failure surface that NETEVO's Implicit Authority Cascade (IAC) names: set clear ownership and guardrails before launch, unique risks of this emerging technology, scaling to production. At the public-sector level it gives practitioner expression to the same distinction the paired Bounded SaaS / Unbounded Agents framing draws, complementing ISO/IEC 38507, which treats the same governance question at the international-standard level.

The first AU instrument purpose-built for agentic deployment. Agentic-AI specificity is rare in current Australian regulator output. The Voluntary AI Safety Standard and the Guidance for AI Adoption address AI generally; the EU AI Act addresses general-purpose AI but not agentic deployment specifically; and the AHRC Human Rights and Technology report predates the agentic-AI surface. The NSW guide is the first Australian instrument to treat agent deployment as a governance regime in its own right rather than as a special case of general AI adoption.

Operational pairing with generative-AI and policy-layer instruments. The guide reads alongside the NIST AI 600-1 Generative AI Profile, which addresses the generative-AI side of agentic systems, and the Cedar runtime policy verification language, which addresses the policy layer. Together the three cover the agentic deployment surface from agency-level checklist to model behaviour to machine-enforceable policy.