Government commits $225m to scale secure AI across the APS
The mid-year economic and fiscal outlook commits $225 million over four years from 2025-26 to accelerate AI across the Australian Public Service, with a further $400,000 each year from 2029-30. The focus: expand the GovAI platform and pilot a secure, desktop-ready assistant called GovAI Chat.
Finance Minister Senator Katy Gallagher set a clear goal in November 2025: give every public servant secure, supportive generative AI from their laptop or desktop. The message to agencies is simple-AI is moving from experiments to daily tools.
Where the money goes
- $166m to Finance and the Digital Transformation Agency (DTA) to expand the GovAI platform and design, build, and pilot GovAI Chat.
- $28.5m available immediately to Finance and the DTA for early work.
- $137.9m over two years from 2026-27, subject to a business case and further assessments.
- $28.9m over four years from 2025-26 to establish a central AI delivery and enablement function in Finance.
- $22.1m to the Australian Public Service Commission and the DTA for AI training and a workforce approach to role and skills shifts.
- $7.7m over four years for an AI review committee to advise on high-risk government AI use cases and ensure consistent, responsible deployment.
- $30m to the Department of Industry, Science and Resources to create an AI Safety Institute to advise industry, support agencies, and protect Australians from AI-related risks.
What this means for your agency
- Expect staged access to a secure AI assistant (GovAI Chat) directly from APS devices via the GovAI platform.
- A central delivery and enablement function will emerge-use it to speed procurement, reuse patterns, and reduce duplicate effort.
- Training and workforce support are funded-plan now for role design, uplift pathways, and practical on-the-job usage.
- High-risk projects will face committee review-bake in risk controls early to avoid rework.
Adoption with accountability
Local industry leaders back the direction but question the global scale. "In Australian terms, we would see that as a good budget initiative in AI ... [but] on an international scale, maybe not so much," said Michael Gately, CEO and co-founder of Trellis Data. "We're not trying to compete as a tier one performer ... so I think the initial amount is trying to increase AI adoption across the APS, and I think that's good."
On training, Gately was blunt: funding alone won't deliver productivity. "Everyone gets trained, but not everyone will really know how to benefit from that." He argues for genuine KPIs-outcomes like "reduce unemployment by X%" or "increase throughput of processing defence applications," not soft measures like saving five minutes on an email.
Guardrails and trust
Senator Gallagher confirmed an AI review committee to strengthen whole-of-government oversight and keep deployments consistent and responsible. Trust-from staff and the public-sits at the core of the plan.
The AI Safety Institute will support agencies and industry while the government leans on existing laws and expert advice rather than rushing hard new rules. For complex models, especially foreign-sourced, Gately pointed to the need for public-private partnerships: sovereign companies can provide deep technical knowledge, while the APS focuses on legitimacy, service quality, and public confidence.
Practical steps to start now
- Nominate an AI lead in each business area to own use cases, risk, and delivery with the central function.
- Prioritise 3-5 high-value use cases with measurable outcomes (e.g., case backlog reduction, processing time cuts, service success rates).
- Define outcome KPIs upfront; tie them to program objectives, not task-level time savings.
- Tighten data foundations: access controls, audit trails, retention, and redaction workflows for sensitive inputs/outputs.
- Map risk classes (low, medium, high) and pre-collect documents for the review committee: data lineage, model choice, testing, bias checks, security.
- Stand up a pilot playbook: 8-12 week sprints, A/B tests vs. baselines, human-in-the-loop sign-off, and rollback criteria.
- Update procurement for model/service swaps, exit terms, data residency, and incident response obligations.
- Invest in skills where it matters: prompt patterns, policy drafting support, summarisation, case triage, code assistants, and analytics.
- Plan change and comms: show safe examples, share wins openly, and make it easy for staff to request new use cases.
Context: scale vs. intent
Australia isn't chasing tier-one AI leadership with this package. It's funding disciplined adoption, guardrails, and practical benefits inside government systems. If agencies focus on outcome KPIs and reusable patterns, this budget can pay off fast.
Useful links
Key people
- Senator Katy Gallagher, Minister for Finance and the Public Service
- Chris Fechner, CEO, Digital Transformation Agency
- Michael Gately, CEO and co-founder, Trellis Data
Your membership also unlocks:
