Our Old Computers Could Let AI Hackers In

Abigail Bradshaw, the head of the Australian Signals Directorate, says AI-powered attacks can exploit Australia's outdated technology.

She gave the warning at the Sydney Dialogue summit in Canberra on Monday, 14 September 2026. The Guardian reported her speech under the title "Australia's outdated technology is vulnerable to AI hacking attacks, signals chief says".

Bradshaw said governments and businesses must spend "enormous" amounts to fix old, weak systems. The cost sits with both public and private owners. She gave no dollar figure. The point was the size of the job.

Fixing them will mean turning off systems people expect to work around the clock. Like closing a road to repair it, some old systems cannot be fixed while they keep running.

Bradshaw said the Directorate does not know how many AI helpers are active online. These are programs that can act on their own on the internet. They work at machine speed, outside checked lists.

She said national security agencies should be able to test AI companies' models themselves. That means direct checking, not just trusting what companies say. Agencies would be part of testing the most advanced systems.

Bradshaw led the Directorate's Australian Cyber Security Centre as of September 2022. Then she said cyber-attacks were increasingly common, sophisticated and agile. She became Director-General in September 2024.

Looking at what this means for Canberra, her message is for boards and budget chiefs as well as ministers. Fixing old systems is rarely popular. It costs a lot, causes disruption, and no one notices when it works. Bradshaw says owners cannot keep putting it off.

In my view, the switch-off is the harder political problem. Governments promise 24-hour services. So do businesses. Properly fixing old or unsupported systems often means isolating or shutting them down for a time. Someone has to approve that outage and take the heat. That call sits above the tech team.

The broader context here is checking and access. Operators cannot count the AI helpers in their systems. Governments cannot check the models behind them on their own. For staff, that means tighter buying rules, records of AI tool use, and deals that allow model testing. None of that is cheap. It is now core national security work.