For all the boardroom enthusiasm about artificial intelligence, a new piece of research suggests the technology is arriving faster than the controls meant to keep it safe. According to a DigiCert survey reported by Australian Cyber Security Magazine, 81 per cent of Australian organisations have already run into an AI-related security incident or identified an AI-related vulnerability in their systems. That is not a warning about some distant threat on the horizon. It is a snapshot of what is happening inside Australian businesses right now.
The number lands at an awkward moment. Australian firms have spent the past two years racing to bolt generative AI onto customer service, coding, marketing and back-office workflows, often before anyone in the security team had a chance to weigh in. The DigiCert figures put a price on that hurry, and they suggest the bill is already being paid in the form of incidents and exposed weaknesses that most organisations would rather not talk about publicly.
What the research actually says
The headline statistic is deliberately broad. It bundles together two different problems: organisations that have suffered an actual AI-related security incident, and those that have discovered AI-related vulnerabilities before anything went wrong. Lumping the two together inflates the figure, but it also tells a useful story. Whether the damage has already been done or merely spotted in time, the overwhelming majority of Australian organisations now have first-hand experience of AI creating security risk rather than just theoretical concern about it.
That distinction matters because the two categories call for very different responses. A discovered vulnerability is, in many ways, a good news story: it means someone was looking, found the gap, and can move to close it. An incident that has already occurred is a different beast, carrying data exposure, reputational damage and, increasingly, regulatory consequences. The survey does not neatly separate how many of the 81 per cent fall into each camp, and that ambiguity is worth keeping in mind before the figure gets quoted as proof that four in five Australian companies have been breached. It does not say that.
What it does capture is scale. When a problem touches this share of respondents, it stops being an edge case and becomes a structural feature of how Australian organisations are adopting the technology. The controls, in other words, are not keeping pace with the deployment.
Two ways to read the numbers
There are broadly two schools of thought about a figure like this. The first, favoured by vendors and many chief information security officers, treats it as a straightforward alarm bell. On this view, AI has vastly expanded the attack surface: models can be manipulated through prompt injection, training data can be poisoned, machine identities and API keys proliferate faster than anyone can track them, and staff routinely paste sensitive material into tools their employer never approved. The 81 per cent, in this reading, is the predictable result of enterprises sprinting ahead of their own governance.
The second reading is more sceptical of the framing. Security surveys are frequently commissioned by companies that sell security products, and a scary top-line number is good for business. Broad definitions, self-selecting respondents and the conflation of incidents with mere vulnerabilities can all push a figure upward. None of that makes the underlying concern fake, but it does argue for treating the precise percentage as directional rather than gospel. The honest interpretation sits somewhere in the middle: AI is genuinely introducing new classes of risk, and some of the reporting around it is calibrated to sound as urgent as possible.
Both readings, notably, point to the same practical conclusion. Whether the true figure is 81 per cent or something lower, the sensible move for an Australian organisation is to assume it has AI-related exposure it has not yet mapped, and to go looking for it.
Why this matters for Australia
Australia is a revealing test case for this problem. Local enterprises have been among the more enthusiastic adopters of generative AI in the region, yet the country’s cyber security workforce remains stretched thin, and many mid-sized firms lack a dedicated security function at all. That combination, high adoption and limited defensive capacity, is exactly the environment in which AI vulnerabilities go unnoticed until they become incidents.
The regulatory backdrop sharpens the stakes further. The federal government has spent much of the past year moving toward firmer expectations around AI, from the standing up of a national AI office to closer scrutiny of automated decision-making in agencies like Centrelink. The mandatory notification regime under the Security of Critical Infrastructure Act and the Privacy Act’s breach rules mean that an AI-related incident is not just an internal embarrassment; for many organisations it is a reportable event with legal weight. A survey suggesting most organisations have already brushed up against such risk implies a wave of disclosure obligations that many boards have not fully reckoned with.
There is also the shadow AI problem, which Australian enterprises have flagged repeatedly through 2026. Staff adopting unsanctioned tools create precisely the untracked exposure the DigiCert figures describe, and no amount of perimeter security helps when the risk walks in through a browser tab. The research reinforces a message local security leaders have been pushing: visibility, not prohibition, is the first order of business.
What happens next
The likely response falls into a few familiar buckets. Expect more Australian organisations to extend their existing identity and certificate management, DigiCert’s core business, to cover the machine identities and API connections that AI systems depend on. Expect governance frameworks to catch up, with AI usage policies, approval processes and monitoring becoming standard rather than aspirational. And expect the market for AI security tooling, already crowded, to grow noisier as vendors compete to be the answer to figures like this one.
The harder work is cultural. Closing the gap between adoption and control means slowing down just enough to ask who approved a given tool, what data it touches and who is accountable when it fails. For a business community that has treated AI adoption as a race, that is an uncomfortable ask. The DigiCert numbers, whatever their precise accuracy, suggest the cost of not asking is already being felt across most of the Australian economy.
Sources: Australian Cyber Security Magazine.
