The Australian Public Service generative AI policy framework, refined through 2024 and 2025, sits in a steady operational phase in 2026. The policy direction is reasonably clear. The implementation across agencies is uneven, and the gap between the policy intent and the practical experience inside individual APS workforces is the more interesting story.

The policy position

The APS generative AI policy framework in 2026 mandates that agencies have an AI use policy, that staff using generative AI tools for substantive work follow approved use patterns, that personal information and sensitive Commonwealth information not be entered into non-approved tools, and that AI-assisted work products be appropriately reviewed. The Digital Transformation Agency continues to coordinate cross-agency activity, and the Department of Finance has continued to provide whole-of-government procurement frameworks for AI tooling.

What agencies are actually doing

Most APS agencies have a workable AI use policy in place in 2026. Most have approved tools available to staff, typically a Microsoft Copilot or equivalent enterprise product, sitting within the agency tenant boundary. Most have done at least baseline staff training. The compliance with the policy is reasonable in most agencies and patchy in some.

The variation across agencies is meaningful. Larger central agencies have invested in dedicated AI capability teams, structured use case pipelines, and routine measurement of AI assistance value. Some smaller agencies have implemented the minimum compliance package and have done little active development of AI use cases. Several agencies sit in the middle, with motivated individual teams pushing forward and limited central coordination.

The use cases that have stuck

Across the APS, the use cases that have moved into routine production are familiar: drafting and editing of routine correspondence, summarisation of long documents for ministerial briefings, basic research and synthesis tasks, and code generation for technical staff. The applications are useful, the productivity gains are real, and the failure modes are well-understood.

The use cases that have remained difficult are the same as in the private sector. Anything involving sensitive policy advice, anything involving security-classified information, anything where the AI output requires substantive subject-matter validation by an expert who is not available. These remain human-driven workflows with marginal AI assistance.

The procurement question

The whole-of-government procurement framework for AI tooling has streamlined the path for agencies to engage approved AI vendors. The bench of approved suppliers has expanded through 2025 and 2026. Pricing for enterprise AI tooling under the APS framework has been competitive and the contracting overhead has reduced.

The procurement of specific custom AI capability — agency-built or vendor-built AI applications addressing specific operational needs — remains more variable. The standard procurement processes apply, and the timelines for substantive AI capability builds are not fundamentally different from any other technology procurement.

The workforce implications

The APS workforce conversation around generative AI has matured. The fear-driven discussion of 2023 — will AI replace public servants — has largely been replaced by the operational discussion of how AI assistance changes individual roles. The roles most clearly affected in 2026 are corporate functions where AI handles a meaningful share of the routine work: communications drafting, basic legal research, internal training material preparation, technical writing.

The substantive policy and operational work of the APS is being supported by AI but not displaced. The shift in skills required across the APS workforce is real and is being addressed through structured training programs in several large agencies.

The 18-month outlook

Expect the consolidation of agency AI policies and AI capabilities to continue. Expect the use case pipeline in larger agencies to continue maturing. Expect the gap between leading and lagging agencies to remain. Expect the procurement framework to continue evolving as the AI vendor market evolves.

The APS in 2026 is past the early adoption phase of generative AI and into the operational and workforce phase. That is a healthier place to be than where the conversation sat 18 months ago.