AI is now financed across almost every line of the Australian defence budget. It is not, however, governed across them. The 2026 National Defence Strategy (NDS) commits about A$425 billion in capability investment by 2035–36, and names AI as a force multiplier in domains from undersea warfare to cyber. Australia’s strategy of denial requires a force that can see further and decide faster, and AI makes that possible at scale. Yet the governance architecture meant to sustain that capability has not been scaled to match.

Three measures would help close that gap: thresholds that slow AI-assisted decision cycles at key escalation points, auditable decision-support for high-stakes outputs, and a clear position on the limits of AI in Australian defence decision-making.

The financing pattern is what makes AI integration in the Australian Defence Force (ADF) structural rather than peripheral. The 2024 Integrated Investment Program (IIP) committed A$8.5 billion to A$11 billion to enterprise data, and information and communication technology – the digital backbone that allows AI to operate across the force. Its 2026 update adds A$14 billion to A$19 billion for theatre command and control, A$27 billion to A$38 billion for space and cyber, and a dedicated line of A$12 billion to A$15 billion for uncrewed systems. The Advanced Strategic Capabilities Accelerator takes prototype capability into service, with up to A$4.3 billion by 2035-36. These figures describe a force whose effectiveness rests on data and algorithms.

Three characteristics of that build-up sit uneasily with current governance arrangements.

The first is the problem of decision time compression. The Joint Air Battle Management System (A$1.99 billion to A$3.15 billion, 2023–2031), the Distributed Ground Station Australia project (A$1.33 billion to A$1.99 billion, 2024–2031) and the wider OneDefence data program are designed to compress the time between detection and response – which is exactly what denial requires. The problem is, however, that the same compression that is an asset in routine operations becomes a liability in a crisis.

In a regional crisis, such as a Taiwan flashpoint or a standoff in the South China Sea, accelerated alerts can build momentum faster than diplomacy can absorb. Leaders are asked to make consequential decisions about readiness and posture before the intelligence picture is complete. Escalation, then, can emerge from speed and uncertainty rather than from deliberate choice. To slow this down, the government should build thresholds into AI-assisted decision support – points at which the system defers to human review before a recommendation is acted on. The US Department of Defense already requires senior level review for autonomous weapons systems that select and engage targets, and NATO’s Principles of Responsible Use for AI in defence include similar oversight requirements.

The second uneasy characteristic is the erosion of the human role in decision making. AI systems are probabilistic and often opaque, and operators working under pressure tend to defer to their outputs – a dynamic known as automation bias. The human role drifts from assessing evidence to confirming a recommendation. That tendency sits awkwardly against the legal requirement for informed human judgement in targeting under Article 36 of Additional Protocol I to the Geneva Conventions and is exacerbated by the operational pressure to take humans out of the loop in pursuit of speed.

The Defence AI Centre, established in 2024, and the RAS-AI Strategy 2040 are sensible foundations, but defence governance ordinarily lags capability integration. The institutional rules for using AI under these pressures are still being written, and Canberra needs to establish a mechanism for auditing high-stakes decisions that are made with AI assistance.

The third difficulty is misperception: how rivals read a capability matters as much as what it does. The Ghost Shark autonomous undersea vehicle, part of the A$94 billion to A$130 billion allocated to undersea warfare in the IIP, and the Integrated Undersea Surveillance System (A$5.6 billion to A$8.3 billion, 2025–2040) – a hybrid network of fixed and mobile sensors – both extend Australian undersea awareness into areas that nuclear-armed competitors consider essential to survivability of second-strike capability. A Chinese naval commander observing that activity during a crisis cannot easily see whether it is oriented towards conventional denial or towards tracking nuclear platforms.

The same ambiguity attaches to AUKUS Pillar II, which accelerates Australian access to AI alongside the undersea, cyber and quantum domains. The mechanism is fear and expectation: when rivals interpret AI-enabled awareness as compressing their options, they may escalate to restore room for manoeuvre. Canberra can and should choose to reduce that ambiguity by refraining from automating certain decisions – and signalling that choice. The United States and China have already affirmed that final decisions in nuclear command and control will remain with humans, at least on paper. A similar commitment from Australia would signal that AI will not be delegated final authority, and, eventually, reduce threat inflation by adversaries.

None of this is an argument against the ADF build-up. AI integration needs to happen. The question is whether governance can keep pace. My policy brief published by the Asia-Pacific Leadership Network in April describes how AI now sits across the ADF and sets out how the three measures described here could work in practice.

This article draws on the author’s policy brief for the Asia-Pacific Leadership Network on AI in the Australian Defence Force.