At Government Innovation Week in 2025, Minister Katy Gallagher launched the AI Plan for the Australian Public Service. A recurring question in the conversations around that event was how to give public servants the confidence and tools to use AI while protecting the people who rely on government services.
A year on, the conversation has moved forward.
In her keynote address at this yera DTA's "AI Government Showcase", the Minister described progress across the APS: more than 110,000 employees have completed foundational AI training, agencies have appointed Chief AI Officers, and GovAI Chat is being tested across government. She also shared examples of AI being used to transcribe oral histories, help taxpayers identify potential errors and support Services Australia staff to find and interpret information.
Those are encouraging signs. They show a public service building capability and finding practical uses for the technology.
The next test is harder: can government turn that activity into services that are easier to use, more reliable and more responsive to the people who need them?
That requires agencies to look beyond access to tools and examine the work around them.
Start with a problem people recognise
The strongest examples in the Minister’s post have a clear purpose.
At the Australian Taxation Office, AI models in MyTax provide prompts that help people identify and correct potential errors before lodgement. The National Library of Australia is using AI to assist with transcribing oral history recordings. Services Australia is trialling AI to help staff navigate and interpret content more efficiently.
These are different applications, but each starts with a recognisable task. A taxpayer needs to complete a return. A library needs to make a collection more accessible. A public servant needs to find the right information.
That is a useful discipline for any agency considering where to invest. Start with the service, process or decision that needs to improve. Then ask whether AI is the right tool, what safeguards are needed and how the result will be measured.
For a service team, the measure might be whether people get an accurate answer sooner. For an internal team, it could be whether staff spend less time searching for information and more time helping people. In both cases, speed needs to sit alongside accuracy, accessibility and the ability to challenge an outcome.
An agency can have high uptake of an AI tool and still leave the underlying service unchanged. The value comes when the work gets better.
Make reuse part of the job
Government agencies often face similar problems, even when their responsibilities are different. They manage large volumes of information, respond to public enquiries, interpret rules and work within complex systems.
The opportunity is to learn from solutions that already exist rather than have every agency start from scratch. GovAI Exchange is intended to help agencies share what they have built, while GovAI Hosting gives them a way to test ideas without setting up their own infrastructure.
Reuse should extend beyond the technology itself. An agency considering a similar application also needs to understand how the first team assessed risk, tested accuracy, consulted staff and handled exceptions. Sharing those lessons can save time and help smaller agencies that may have less capacity to experiment independently.
That does not mean copying a solution without checking whether it fits. Different agencies handle different information and serve different communities. A tool that works well for one task may need changes before it is suitable elsewhere.
The practical question is: what can be reused safely, and what needs to be adapted? Making that easier to answer would help government get more value from work it has already funded.
Give people room to learn, with clear boundaries
Foundational training is an important start. The next step is helping public servants apply that knowledge to their own work.
Staff need to know when an approved tool is appropriate, what information they can use, how to check an answer and when a person needs to step in. Managers also need time and support to redesign workflows with their teams, rather than simply adding a new tool to an already busy day.
At the AI Government Showcase, Minister Gallagher acknowledged that public administration is risk-averse for good reason: when a major programme goes wrong, many people can be harmed. She described the challenge as finding room for innovation and experimentation with safeguards against harm.
That balance needs to be visible in day-to-day decisions. Low-risk uses can provide a place for teams to build experience. Higher-risk applications need stronger scrutiny, clear accountability and appropriate human oversight. Staff should be able to raise concerns and report problems without feeling that doing so will slow progress.
Trust is built through that practical work, especially when agencies explain clearly what a system does and how people can seek help if something goes wrong.
Deal with the systems AI depends on
A useful AI tool still depends on reliable information and the systems around it. If guidance is outdated, data is fragmented or a service has multiple conflicting sources of truth, an AI assistant may make the problem easier to encounter rather than solve.
This matters for the public as well as government staff. People already use chatbots to ask questions about government services. If the underlying information is difficult to find or inaccessible, they may receive an incomplete or incorrect answer. The Minister raised this challenge in her showcase address.
Agencies will need to improve the quality and accessibility of their information, be clear about what can be shared, and maintain the systems that support essential services. Some legacy systems will need significant investment. Others may be retired or improved in stages.
There is no easy way around those choices, particularly with competing demands on public money. The priority should be the systems and information that have the greatest bearing on service quality, security and people’s ability to get what they need.
Create room for Australian capability
The Minister also highlighted the government’s commitment to supporting sovereign AI capability and changes to procurement rules developed with the SME sector and Australian companies.
For government, the benefit of a broader supplier base is choice. Agencies can draw on local expertise, test different approaches and reduce their dependence on any one provider. For Australian businesses, it creates an opportunity to solve public sector problems and build capability that can be applied elsewhere.
That opportunity will be strongest when agencies describe the outcome they need, make procurement requirements proportionate to the risk, and give smaller suppliers a fair chance to demonstrate what they can do. Suppliers, in turn, need to show how their products fit public sector obligations, integrate with existing systems and deliver measurable value over time.
Government should also understand the practical consequences of its technology choices: who controls the data, what it costs to operate and maintain a service, and how difficult it would be to change providers later.
Measure the result for Australians
The progress described by Minister Gallagher provides a useful base for the next phase. More public servants have access to training, leadership and tools. Agencies are beginning to share examples of work that could help others.
The opportunity now is to connect those efforts to outcomes people can see: clearer information, fewer avoidable errors, more accessible services and public servants with more time to help.
The most useful question for the next government AI project is a simple one: what will be easier or better for the person using the service, and how will we know?
Want to learn more about how governments are using AI globally? Become a member and join (for free) to access content and case studies from around the world.
Published by
Help your peers
Share what you've learned with fellow public servants