For the better part of two years, artificial intelligence has been pitched to Australian boards as the answer to almost everything, and cyber security has been no exception. Vendors talk about autonomous threat hunting, self-healing networks and models that spot an intruder before a human analyst has finished their morning coffee. The pitch is seductive, especially for organisations that have spent the past decade watching their attack surface balloon while skilled defenders stayed scarce. The trouble, as a growing chorus of experts keeps pointing out, is that technology bolted onto a shaky foundation tends to magnify the cracks rather than paper over them.
That is the tension running through the latest guidance from the nation’s cyber authority. In its Annual Cyber Threat Report 2025-26, the Australian Signals Directorate encourages organisations to lean into AI for defensive work, recognising that attackers are already using the same tools to move faster and at greater scale. Yet the report stops well short of presenting AI as a silver bullet. The practical advice still circles back to the unglamorous disciplines that determine whether an organisation survives a bad day: knowing what assets you own, patching them promptly, controlling who can log in, and having a plan for when something goes wrong.
The news
Analysis published by ASPI’s The Strategist sharpens the point, arguing that scattering what it memorably calls AI pixie dust around an organisation does not automatically deliver cyber defence. The real question, the piece contends, is not whether you have adopted AI but how much genuine capability it adds once you account for the effort, the data it needs and the new weaknesses it introduces. An AI system is only as good as the telemetry feeding it, the people interpreting its outputs and the processes that act on its alerts. Strip those away and you have an expensive dashboard that produces confident noise.
The directorate, led by Director-General Abigail Bradshaw, has consistently framed the threat environment as one where the fundamentals carry the most weight. Its long-running Essential Eight model of mitigation strategies, covering things like application control, multi-factor authentication and timely patching, remains the baseline it urges every organisation to reach before chasing anything more exotic. The message in this year’s report is that AI can help an organisation do those fundamentals better and faster, but it cannot substitute for having them in place.
Two ways to read it
Optimists in the security industry will argue that AI is already earning its keep. Modern defenders are drowning in alerts, and machine learning has become genuinely useful at triaging that flood, correlating signals across sprawling networks and flagging the handful of events that warrant a human’s attention. Used well, it compresses the time between an intrusion and its discovery, which is often the difference between a contained incident and a front-page breach. For stretched teams, that is not hype, it is survival.
The sceptical reading, which the ASPI analysis leans towards, is that AI arrives with its own baggage. Models can be poisoned, prompts can be manipulated, and the rush to deploy generative tools has opened fresh avenues for data to leak out of organisations that never properly understood where their information was going in the first place. There is also a governance problem. A board that believes it has solved cyber security by buying an AI product may invest less in the tedious, continuous work that actually reduces risk. In that sense, the pixie dust can be worse than nothing, because it breeds a false sense of safety.
Both readings can be true at once, and that is really the directorate’s point. AI is a powerful amplifier. Pointed at a mature security program it makes good defenders better. Pointed at an immature one it amplifies confusion and gives attackers, who are using the same technology, more surface to exploit.
What it means for Australia
The Australian stakes here are concrete rather than abstract. The directorate continues to respond to thousands of cyber incidents a year, and the cost of cybercrime to local businesses runs well into the billions, with small and medium enterprises bearing a painful share. These are precisely the organisations most likely to be sold AI as a shortcut, because they lack the in-house expertise to tell a transformative capability from a marketing slide. Many do not yet have reliable asset inventories or consistent patching, let alone the data pipelines a serious AI defence tool would require to function.
There is a sovereignty dimension too. Australia has been pushing to build domestic AI capability, from sovereign compute through to local models, and cyber defence is one of the areas where that ambition meets reality. Relying on offshore AI security platforms raises questions about where sensitive telemetry is processed and who can see it. For critical infrastructure operators now bound by the Security of Critical Infrastructure regime, those questions are not academic, and regulators will expect answers that go beyond a vendor’s assurances.
The skills shortage compounds all of this. Australia simply does not have enough experienced security professionals, which is part of why AI looks so appealing as a force multiplier. But a multiplier needs something to multiply. An organisation with no internal capability cannot safely hand its defence to a model it does not understand, because someone still has to decide what the model should watch, judge whether its outputs make sense and act when the alarm sounds at two in the morning.
What’s next
Expect the directorate and its partners to keep hammering the same theme through the coming year: adopt AI, but earn the right to by getting the basics right first. Practically, that means Australian organisations should treat any AI security investment as a layer on top of a working program rather than a replacement for one. The sensible sequence is to secure identities, maintain a current picture of assets, patch quickly, segment networks and rehearse incident response, then look at where AI can accelerate those disciplines.
Boards, meanwhile, would do well to ask harder questions of the vendors parading through their offices. What data does this tool need, where does it go, and what happens when it is wrong? How does it fail, and who notices? The organisations that come through the next wave of threats in good shape will not be the ones that adopted AI earliest. They will be the ones that built sturdy defences and used AI to make them sturdier still.
Sources: ASPI The Strategist
Comments 1