The following blog is an edited version of Dr Paul Hubbard’s keynote address at the ‘Future Ready APS’ event hosted by the Mandarin and the Hatchery in Canberra on 17 June 2026
“AI – are we there yet?”
AI is already in our inboxes and already in the hands of the citizens we serve. The APS plan is clear: every APS officer, every agency, will have the tools, the capability and the guidance to use AI safely and effectively to advance the public good.
This isn't "I get to do AI because I happen to have a good data team." It's for all of us. As my kids would say, it's a "you" problem, meaning each and every person reading this. What transformation looks like depends on your context — individual, organisational and service.
Literacy is not the finish line — it's the start
We need to understand the basics of what AI is and how large language models work. A lot has been invested in that through the APS Academy, through GovAI, through GovAI Chat. GovAI Chat is now connected to more than 30 agencies, with over 1,500 staff and growing every day.
The tools are here.
The chance to get hands-on is here.
Literacy is a foundation, not a destination.
If we teach people to prompt well but never tell them why, we've missed the point entirely. We don't need 200,000 public servants to become AI engineers. We do need every public servant to be able to exercise professional judgement about when, why and under what conditions AI should be used in the domain they serve.
Understanding the tools is the necessary foundation; understanding the context is what lets us understand the consequences.
The crevice becomes a chasm
This is where the capability gap between us and the public becomes a crevice and the crevice becomes a chasm because the technology the public is already using grows more capable every day.
At the risk of sounding like AI slop: change is happening faster than ever.
The recent collaboration between Anthropic and other technology companies showed that a frontier model could find and write exploits for the kinds of software we all rely on every day. The same project then worked to patch that software before deployment. That capability did not exist six months ago. So if your latest understanding of AI was six months ago, you've missed something like two decades' worth of change.
Even within the last few weeks, a new AI model was released. I was very excited to be using it Wednesday, Thursday, Friday and then on the Saturday I got a message saying I couldn't use it anymore because I'm not a US citizen. (As it turned out, it was pulled for everyone because working out who is and isn't a US citizen is harder than it sounds.)
The point is: the sort of change we used to metabolise over decades now arrives in months, and increasingly in weeks.
This poses a real challenge for anyone working at the intersection of technology and government, because our instinct is to ask: how do we control this technology, how do we manage these risks? But it isn't only the technology inside our agencies that matters, it's the technology everyone is using, and it's moving in ways that are unpredictable and exponential.
Not, mostly, a technological challenge
So what can we do? For the most part, those of us working in government aren't coders, which is good, because this is not primarily a technological challenge. There are genuine technical assurance questions, of course. The role of a Chief AI Officer in an APS agency is not the same as a Chief Information Security Officer or a Chief Data Officer. It's about how we transform public service in a way that builds public trust.
Not "how do we deliver services while maintaining trust?" but "how do we deliver services that build trust?"
This takes me back to human-centred design and engaging the community in the process. You cannot do safe and responsible AI that serves citizens if citizens haven't been part of the process.
Safe and responsible AI is not just a technological standard. The standards are part of it, but they're not the whole thing. If I build a tool I think someone will like, but I never ask them and never bring them along on the journey, then it's perfectly reasonable for them to be mistrustful rather than curious.
Some good news
We know trust is important. There isn't a lot of good news in the world today, so let me share some. Trust in government services rose last year to 62 per cent, according to the APSC's Trust in Australian Public Services report.
And here's the telling part. When people were asked why they trusted, they didn't talk about technology. They didn't say government got good at tech, or that it rolled out a co-pilot. They talked about ease of access to services and information. They talked about faster response times, better transparency in decision-making, and being genuinely helped.
The public isn't wondering whether you used Opus or Sonnet or Haiku. They care whether the interaction was faster, fairer and less confusing.
Where collaboration happens
The AI CoLab is a cross-sector initiative that came out of the APS Reform agenda, funded by the APS Capability Reinvestment Fund. Since it was established, it has had around 1,330 participants across roughly 120 workshops. About half come from government; Commonwealth, ACT and New South Wales with around 25 per cent from the private sector, 20 per cent from the community sector, and roughly 5 per cent from academia.
It's a real space where co-design can happen: where we can talk about the issues that matter, surface the open questions none of us can answer alone, and try to make sense of them together.
The final mile of AI adoption is 90 miles long
Technical brilliance does not make you ethically literate, and it doesn't necessarily help the public. We see AI systems that are engineering marvels but completely detached from our context. In that sense, Silicon Valley is almost a different planet. The technology is amazing; translating it into our context is the work and that work is ours.
The final mile of AI is about 90 miles long, and that's our job in the APS. It belongs to the people who understand local communities, power dynamics, vulnerability and historical context — the people who understand what something like Robodebt meant to people, and the lessons we never want to repeat. The arts graduates, the anthropologists, the linguists, the generalists in the public service — this is why AI is everybody's problem. And the APS already has these people. Code is now easier; the real value is our people and the context they apply.
One thing to take away
None of us has the answers individually. Responsible AI comes from genuine collaboration, co-design and a human-centred approach. That's not a nice-to-have or a box to tick at the end — it's foundational to how we do safe and responsible AI.
The technology is malleable.
It can be reshaped.
Yes, we will be shaped by the tools we use — but we have the opportunity, now, to shape those tools for the better.