Australia’s Defence Force is spending big on artificial intelligence, and it is doing so almost everywhere at once. From maritime surveillance and targeting to logistics, intelligence analysis and back-office administration, AI has quietly become a line item in nearly every major capability program. The problem, according to a new analysis, is not that the money is flowing. It is that the guardrails have not been built to match.
Writing for the ASPI Strategist, the Australian Strategic Policy Institute’s commentary arm, the argument is blunt: AI is now financed across almost every part of the defence budget, but it is not governed across them. Investment has raced ahead of the frameworks meant to keep it accountable, auditable and safe, leaving Defence in the awkward position of deploying capabilities faster than it can explain how they will be controlled.
The scale of the commitment
The context is the 2026 National Defence Strategy, which sets out roughly A$425 billion in capability investment through to 2035-36. That figure covers the full sweep of Australia’s rearmament, from nuclear-powered submarines under AUKUS to long-range strike, integrated air and missile defence, and the digital backbone that ties it all together. AI is not a separate bucket in that plan. It is woven through the programs, embedded in sensors, decision-support tools, autonomous systems and the data pipelines that feed them.
That diffusion is precisely what makes governance hard. When AI sits inside dozens of individual projects rather than in one dedicated program, there is no single owner, no single budget line and no single set of rules that applies to all of it. Each project team makes its own calls about testing, assurance and human oversight. The result is a patchwork, where the standard of governance depends on which program you happen to be looking at rather than on any consistent, force-wide policy.
Why the gap matters
The stakes here are different from those in the commercial world. When a bank or a retailer gets an AI system wrong, the consequences are financial or reputational. When a military system errs, the consequences can be lethal, legally fraught and strategically damaging. Questions about how an autonomous system selects a target, how much a human operator can realistically intervene, and who is accountable when an algorithm gets it wrong are not academic. They sit at the heart of international humanitarian law and of Australia’s obligations as a responsible military power.
Defence has not been silent on this. Australia has signed up to international statements on responsible military use of AI, and the department has published its own thinking on ethics in autonomous systems through work led by the Defence Science and Technology Group. The concern raised in the ASPI analysis is not that principles are absent. It is that principles articulated at the strategic level have not yet been converted into the hard machinery of governance: mandatory assurance standards, testing and evaluation regimes, audit trails, and clear accountability lines that apply uniformly across every AI-enabled program.
Two ways to read it
There is a genuine tension underneath this debate, and it is worth stating both sides fairly. The case for moving fast is strong. Australia’s strategic environment has deteriorated, the government has been explicit that the country no longer enjoys the luxury of long warning times, and adversaries are investing heavily in AI-enabled warfare of their own. In that light, waiting for perfect governance before fielding useful capability could itself be a strategic risk. Speed has value when the threat is real and growing.
The counter-case is that governance is not a brake on capability but a condition of it. Systems that cannot be trusted, verified or explained are systems commanders will hesitate to rely on in a crisis, and allies will hesitate to plug into. Poor assurance can also become a single point of failure, whether through an adversary poisoning training data, an unnoticed bias in a targeting model, or a brittle system that behaves unpredictably outside the conditions it was tested in. On this reading, skipping the governance work does not buy speed. It defers a bill that comes due at the worst possible moment.
The Australian stakes
For Australia, the issue lands squarely in the middle of the sovereign AI conversation that has been running through Canberra and industry all year. The country wants to build and control critical AI capability at home rather than rent it from abroad, and Defence is one of the most demanding customers for exactly that kind of sovereign capability. Yet sovereignty over hardware and code means little without sovereignty over assurance: the ability to independently test, certify and stand behind the systems the ADF fields.
There is an allied dimension too. AUKUS Pillar Two explicitly covers advanced capabilities including AI and autonomy, and interoperability with the United States and United Kingdom will depend on shared confidence in how these systems are governed. If Australia’s frameworks lag, it risks becoming a consumer of others’ standards rather than a shaper of them. That matters for a country that has spent considerable diplomatic capital positioning itself as a credible, rules-respecting middle power on emerging military technology.
The workforce question sits alongside it. Governing AI at scale requires people who understand both the technology and the law of armed conflict, and those people are scarce and expensive. Defence is competing for the same talent as banks, data-centre operators and start-ups, at a time when the broader Australian economy is already short of AI expertise. Building the assurance capability the ASPI analysis calls for is as much a hiring and training challenge as a policy one.
What happens next
The practical fix being pointed to is not to slow the spending but to build the connective tissue around it. That means a coherent, force-wide governance framework rather than program-by-program improvisation, clear ownership of AI assurance inside Defence, and standards that travel with a capability from procurement through to operational use and eventual retirement. It also means transparency deliberate enough to survive parliamentary and public scrutiny, given how much taxpayer money is involved.
None of that is glamorous, and none of it will generate the headlines that a new drone or missile does. But the argument running through this analysis is that the governance layer is what turns raw capability into something a commander can actually trust and a government can defend. With A$425 billion committed and AI already threaded through the plan, the window to build that layer is open now. The warning is that it will not stay open indefinitely, and that catching up later tends to cost far more than getting it right at the start.
Sources: ASPI Strategist.
