The Australian Public Service procurement practice for AI services has shifted meaningfully through 2025 and into the first half of 2026. The combination of evolving AI policy direction, post-PwC probity tightening, and the practical experience of running AI procurements through the past 18 months has reshaped how APS agencies approach AI service buying.

What is different in May 2026 compared to early 2024:

The AI assurance documentation expectation has hardened. APS agencies running AI procurement in 2026 are routinely asking for documented AI risk position, model card information, training data provenance position, and operational assurance plans. Suppliers without structured AI assurance documents are increasingly being excluded at the early procurement stages.

The reference customer expectation has tightened. AI procurement assessments in 2026 are placing more weight on supplier reference customers than equivalent procurements in 2023. The reference checks are also more detailed — focused on production AI delivery outcomes, on AI operational discipline, and on the customer-facing service model around the AI capability.

Sovereign hosting and onshore delivery is being scored more heavily. The pattern has been clear through 2025 — APS agencies are routinely requiring onshore hosting of AI workloads dealing with government data, Australian-cleared personnel on the delivery team, and contractual provisions confirming the onshore posture. This is not new but the scoring weight is heavier in 2026 than it was 18 months ago.

The IRAP-aligned environment is the default expectation. Most APS AI procurements are now structured around IRAP-aligned environments at the relevant security classification. Suppliers proposing AI capabilities without a current IRAP-aligned operating environment are operating at a significant procurement disadvantage.

What patterns are emerging in the procurement structures:

Bounded-scope agile delivery is the dominant procurement pattern. The 2018–22 fashion for large multi-year AI contracts has been replaced by bounded-scope agile engagements with optional extension phases. The typical engagement structure is a 4-8 week discovery phase, followed by a 12-16 week build phase, followed by a managed service or capability transfer phase.

Multi-supplier panel arrangements are being used more flexibly. APS agencies are running AI procurements through existing technology panels rather than running standalone procurements where possible. The panel arrangements have been updated to include AI service line items at most major agencies through 2024–25.

Skills-based hiring through panel arrangements is being combined with capability-led engagements. The pattern at well-organised AI programs in APS agencies is a small core team of APS staff working with a panel-supplied specialist team and an external capability vendor providing platform and engineering depth. The combination of the three has worked better than pure outsourced delivery or pure in-house delivery.

What the supplier conversation looks like in 2026:

The major consulting firms remain active in APS AI procurement but their share of delivery has reduced from the peak years. The post-PwC environment has been more difficult for firms with significant policy advisory work alongside delivery work and the procurement assessments have been more cautious about awarding delivery work to firms with adjacent advisory engagements.

Mid-tier specialist Australian technology firms have grown materially through 2024–25 on APS AI work. The procurement criteria around production AI delivery experience, AI engineering depth, and clean policy advisory boundaries have suited smaller specialist firms.

Microsoft-aligned suppliers are well-positioned given the Azure and Copilot footprint across the APS. Suppliers with established Microsoft AI delivery practices are positioned to compete in most APS AI procurements that are running on Microsoft technology.

Operational notes for AI suppliers planning APS work through the rest of 2026:

The probity and conflict-of-interest position needs to be documented up-front. Suppliers with active policy advisory engagements in adjacent areas should be transparent about the position from the beginning of the procurement. The assessments are sophisticated and the procurement teams are looking for these issues.

The AI assurance position needs to be a written, structured document. Suppliers that can present a structured AI assurance package — risk position, model documentation, training data provenance, operational governance — are getting through assessment faster than suppliers answering each question separately as it arises.

Past performance documentation needs to be specific. Generic past-performance statements are not sufficient. The procurement assessments are looking for specific named projects, specific outcomes, and reference customers willing to engage with the assessment process. Suppliers with strong reference customers are at a meaningful advantage.

Onshore delivery commitments need to be operationally credible. Procurement assessments are increasingly skeptical of onshore-on-paper arrangements where the actual delivery model relies on offshore resources. Suppliers should be transparent about the delivery model and structure proposals accordingly.

For AI suppliers and APS procurement teams working through the rest of 2026, the practical read is that the procurement practice has matured and the expectations are higher than they were 18 months ago. The procurements are running more carefully, the assessments are deeper, and the supplier preparation required to compete successfully has increased. That is broadly a good thing for the long-term quality of APS AI delivery and the operational maturity of the supplier ecosystem.

The next 12 months will likely bring continued operationalisation of AI assurance practice across the APS, continued maturation of supplier capability assessment, and a steady increase in production AI volume across federal departments. The procurement practice will continue to evolve and the suppliers that can keep pace are well-positioned for the next phase.