Old systems, new threats and another very bad night for cheese
First, they came for the cheese. In an overnight cyber attack on 11 October, refrigeration systems at hundreds of supermarkets across France were forced into continuous defrost cycles, with some reportedly reaching 50°C. The result was spoiled food and disrupted business – an illustrative example of the impact of a cyber attack on daily life. Danish manufacturer Danfoss, whose controllers were among those affected, said many affected units had reached end of life or were running outdated firmware.
Then came a second helping. On 2 November, retailers in seven European countries reported another wave of refrigeration incidents, with some units again reaching 50°C. Several companies confirmed cyber attacks, and even an Ikea store in Belgium was affected. It remains unclear who was responsible or whether the incidents were linked to the French attacks.
AI’s involvement in either wave of refrigeration incidents has not been established, but its growing ability to help find and exploit weaknesses is adding pressure to an already difficult task: securing large numbers of older, connected systems. The gap between identifying a vulnerability and protecting the systems exposed to it is becoming harder to manage.
The original ‘cheese hack’ came just weeks after our own Prime Minister publicly confirmed that an OpenAI agent had gained unauthorised access to Services Australia’s Medicare statistics portal during a research task in June. OpenAI says individual patient or client records were not accessed, but the model retrieved internal files and credentials. A task intended to gather public information had crossed the digital equivalent of a locked door.
Then, last month, governments around the world temporarily took websites offline following the disclosure of critical vulnerabilities in PHP — software that underpins a large part of the web. Known as PHP2Shell, the flaws showed how AI could help researchers uncover longstanding weaknesses and develop working exploits in controlled testing. The same capabilities could accelerate attacks on unpatched systems, even as they help defenders find and fix the problems.
Taking key government websites offline to address security flaws might once have seemed extreme, but it now illustrates the difficult choices cyber teams face as the window for action shrinks. The latest Annual Cyber Threat Report from the Australian Signals Directorate (ASD) warns that defenders who once had months to patch now have days — sometimes only hours — with AI further accelerating discovery and exploitation. The challenge lies in turning faster discovery into faster protection.
These incidents have different causes, but together they show how familiar systems can become urgent operational risks. A refrigeration controller, a reporting portal or a widely used piece of web software can carry on doing its job for years. That familiarity offers little reassurance when the ways those systems can be accessed or exploited change.
Those watching Australian government cyber policy will also have noticed a pattern. At the end of September, Home Affairs Secretary Stephanie Foster issued a direction requiring non-corporate Commonwealth entities to complete a legacy technology stocktake and risk management plan by the end of March 2027. The direction also calls for rapid patching of critical vulnerabilities. The legacy technology problem is hardly new, but the direction joins a long-running modernisation task to an immediate security obligation.
An announcement from ASD on 15 October was equally telling: the Essential Eight, its cybersecurity framework, was getting a glow-up. Introduced in 2017 primarily for enterprise IT networks, it is now being expanded through the ASD Essentials framework, with guidance covering enterprise IT, cloud services and operational technology. The Essential Eight remains in place during the transition.
Controls designed for enterprise IT networks do not always map cleanly onto shared-responsibility cloud models. A provider and its customer have different responsibilities, which vary with the service. Organisations need to know what the provider delivers, what they must configure or operate themselves, and what evidence they need that the required safeguards are working.
ASD says the Essential Eight remains effective, and that existing controls and investments can carry forward into the new framework, which matters for organisations that have spent years putting them in place. The new guidance builds on previous work, while asking organisations to examine where protections need to be adapted or extended: a change of framework is no reason to put existing security improvement programs on hold.
Taken together, the government’s stocktake direction and ASD’s new guidance point to a changing approach. Organisations are being asked to understand the systems they rely on, prioritise their most pressing risks and demonstrate that security measures achieve the intended outcomes in their actual environment. The assessment needs to reflect how those systems operate and connect, including where responsibility is shared.
This also means revisiting assumptions about how work happens, who has access and how security decisions are made. An organisation may know that a system needs replacing without having settled who can authorise a patch, restrict access or take a service offline in the meantime, and a replacement program might take months or years. Reassessing those assumptions and addressing the most urgent exposures can begin now.
The timing matters because work now happens across cloud services, browsers and desktops, on corporate and personal devices, and increasingly through AI agents acting on behalf of people, and security policies need to reflect that reality. Organisations need to understand who or what is accessing information, what they are permitted to do with it, and whether those permissions remain appropriate as work changes. The question for leaders is whether their security arrangements still fit the way their people actually work.
Evidence is ultimately the test: organisations should be able to show which controls apply, how they are enforced and what happens when someone, or something, tries to step outside them. Audit trails can show who accessed data and what they did; and testing can show whether safeguards detect, block or contain activity that breaches policy. The findings should inform decisions about what to fix first and who has authority to act. Replacing legacy systems takes time, and demonstrating and improving protection for the systems in use today cannot wait until the replacement program is finished.
Government can strengthen security without rushing to replace proven systems
The Australian Government's directive on legacy technology has placed renewed pressure on...
Why AI is raising the stakes for legacy government IT
Government organisations at every level must understand the risks associated with aging...
What replaces the Essential Eight needs to fix nine years of blind spots
The soon-to-be-retired Essential Eight will need to address the many blind spots the original...
