Source: Trust, attitudes and use of artificial intelligence, University of Melbourne with KPMG, 2025. More than 48,000 people across 47 countries.
The gap
Half of us use it. A third of us trust it.
Sit with those two numbers for a second, because they describe a lot of Australian workplaces right now. People are already using AI. They are just not telling anyone, and they are not sure they should be.
That gap is not stupidity and it is not fear of the future. It is a perfectly sensible response to being handed a powerful tool with no instructions, no ground rules, and no one to ask when it says something that sounds confident and turns out to be wrong.
You cannot close that gap with enthusiasm. You close it by giving people a clear line between what is safe to put in and what is not, and by letting them practise on something that does not matter until it does.
Where the country is heading
Safety is not the brake. It is the enabler.
“The single greatest threat to our AI success is not a shortage of talent, capital or energy, but a shortage of trust.”
That is not a technology company talking, and it is not me. That is Dr Andrew Charlton, Australia's Assistant Minister for Science, Technology and the Digital Economy, speaking at the Australian AI Safety Forum on 7 July 2026.
He made a second point that I think is the one small businesses need to hear. AI is not going to stay in a box marked "tech". It is becoming, in his words, a general-purpose technology, like electricity. It will be in every hospital. Every classroom. Every business. Every defence capability. Every government service.
Nobody asks whether they are "using electricity" any more. They just turn the light on. AI is heading the same way, which means the question is not whether it turns up in your work. It is whether you are ready when it does.
“No country will win the AI race with technology that its own citizens don't trust, and the nations that build safety in from the start will be the nations that capture the prize.”
The Hon Dr Andrew Charlton MP, Assistant Minister for Science, Technology and the Digital Economy. Read the full speech
Watch it yourself
Twenty minutes, from the person setting the policy.
I would rather you heard this from him than from me. It is not technical, and it is easy to follow.
Opening keynote, Australian AI Safety Forum. Published by the AI Safety Forum.
The risks, honestly
It does exactly what you asked for. That is the problem.
I am not going to frighten you, and I am not going to pretend there is nothing here. The honest position sits in between, and it starts with a story the Assistant Minister told, about a boat race.
2016, a video game
Researchers trained an AI to play a boat-racing game and rewarded it with points. The system worked out that the points came from hitting targets along the course, not from finishing the race. So it found a lagoon, drove in endless circles catching fire and crashing into other boats, and scored higher than any human player, without ever crossing the finish line.
We asked for points when we meant winning. It gave us exactly what we asked for.
That was funny. What has happened since is less funny. Last year researchers set the most advanced models the task of beating one of the world's strongest chess engines. One model, finding itself in a losing position, reasoned that its job was to win against a powerful chess engine, “not necessarily to win fairly”. It went and hacked the files holding the positions of the pieces, and forced its opponent to resign.
The 2026 International AI Safety Report, written by more than 100 experts under the Turing Award winner Yoshua Bengio, found that risks once considered theoretical are now turning up in the evidence. Frontier models are showing early signs of deception, cheating and situational awareness. The word for this is misalignment: a system pursuing a goal subtly different from the one you thought you gave it.
None of this means the newsletter your team drafts on a Tuesday is about to hack anything. It means something simpler and more useful to you. These systems optimise for what you actually asked, not what you meant. So the instruction matters, and the person checking the output matters even more. That person is you, and that is not a burden. It is the job.
What is coming next
Nobody breaks a rule, and there is still a traffic jam.
The next thing coming is not a cleverer chatbot. It is AI agents: systems that can understand information, make decisions, and take action to achieve a goal with limited supervision. Not tools you prompt. Things you delegate to.
Picture a city where every business has hired an army of digital assistants. One negotiates contracts, one manages deliveries, one handles customer service. They do not just talk to us any more. Increasingly they talk to each other.
The best analogy I have heard
It is a bit like city traffic. Every driver may be following the rules, and you still get congestion, accidents and gridlock, simply from the way everyone interacts.
Risks can emerge between agents even when nobody has done anything wrong, and no single organisation can see the whole road.
That is a genuinely new kind of problem, and it is why the answer is not one big AI law. It is every regulator who already knows their own sector moving faster. In the minister's words: “That is not fewer rules. That is faster rules, applied by regulators who already understand their sectors.”
Why now
“The window to get ahead of this technology is open now. It will not stay open forever.”
The Hon Dr Andrew Charlton MP, Australian AI Safety Forum, 7 July 2026
Where to go next
Australia is not making this up as it goes.
There are real institutions doing this work, and their material is free and public. If someone tells you nobody is minding the shop, send them here.
AISI
The AI Safety Institute has three goals.
- 1 Analyse and test new AI models and applications. It was testing frontier models with technical partners in its first month, not in year three and not after a review.
- 2 Support regulators and agencies to respond to emerging AI capabilities, risks, harms and trends, so policymakers act on evidence rather than headlines.
- 3 Shape safe AI development, deployment and international governance in Australia's interests.
It works closely with regulators without being a regulator itself. And as the minister put it, “Australian society has a big stake in the development and implementation of AI in Australia.” That includes you.
National AI Centre
ai.gov.au
The front door: guidance, tools and resources to help Australian businesses use AI safely. Plain sections on knowing the risks, preparing your business, and handling the AI tools your people are already using without telling you.
Australian AI Safety Institute
Announced November 2025, operating since early 2026
Australia's own institute for testing frontier AI models and monitoring emerging risks. It began testing frontier models with technical partners in its first month. In the minister's words it “works closely with regulators without being a regulator itself”.
The National AI Plan
Released December 2025
The government's plan for capturing the benefits of AI while keeping Australians safe. If you want to know where the country is heading, and why safety and adoption are being treated as the same job, it is here.
Australian AI Safety Forum
aisafetyforum.au
Where the keynote above was given, at the University of Sydney on 7 and 8 July 2026. The talks are going up as videos, free to watch.
Gradient Institute
gradientinstitute.org
An independent, not-for-profit research organisation in Sydney, working on safe and responsible AI through research, advisory work and education. Not government, not a vendor. Their resources are free.
Australia also sits inside the international network of these institutes, which was renamed in December 2025 to the International Network for Advanced AI Measurement, Evaluation and Science. That is a mouthful, and it matters: it means the testing that catches a model cheating at chess gets shared across borders rather than buried.
What it means for you
Swap the word "safety" for the word "trust".
When a government says AI safety, it means model evaluations, standards and regulators. When I say it to a council team or a small business, I mean something much closer to home. Can your people use this without putting something in it they should not have? Can they tell when it is confidently wrong? Would you be comfortable if a client saw exactly how that piece of work got made?
That is not a technical problem. It is a training problem, and it is a completely solvable one.
Understand what it actually does
Not the hype and not the doom. What the tool is, where the words come from, and why it sometimes makes things up with a completely straight face.
Know what is safe to put in
A clear green, amber and red line for your own information, so nobody has to guess on a Tuesday afternoon with a deadline in front of them.
Stay the human in the loop
The work still goes out with your name on it. AI drafts, you decide. That is the whole deal, and it is the part that keeps your reputation yours.
This page is a practical starting point, not legal advice. If you handle health, financial or personal information, check your own obligations before you put anything into an AI tool.
Why this page exists
Information reduces fear.
I have watched a lot of rooms go quiet when AI comes up. Not hostile. Quiet. People are worried they will look foolish, or that they have already done something wrong with it, or that it is coming for the thing they are good at.
Almost none of that survives contact with actual information. When you know what the tool is, what it does with your words, who is testing it and what the country is putting in place, the fear does not vanish, but it stops running the show. You become harder to frighten and harder to fool. That is not a small thing. Fear and not knowing are what make people vulnerable, and vulnerable people either avoid this technology entirely or hand it far too much.
So that is what this page is for. Not to sell you a course. To make you a little harder to alarm, and a little harder to take advantage of.
New to all of this?
Most teams are. Let us make a start together.
If your team is already using AI quietly and nobody has drawn the lines yet, that is usually the right time to bring me in. Tell me what your team does and I will tell you what I would put in front of them first. No pressure at all.
Start a conversation