Australia’s Signals Chief Warns AI Agents Could Exploit Ageing Infrastructure

Australia’s signals intelligence chief has warned that ageing government and business technology could become increasingly vulnerable as AI agents give cyber attackers the ability to operate at machine speed and scale.

Abigail Bradshaw, director-general of the Australian Signals Directorate, was asked how many AI agents might already be operating across the internet.

Her answer was concise: “A lot.”

Putting a meaningful number on them, she said, was effectively impossible.

The concern is not simply that increasingly capable AI models might discover new forms of attack. Existing weaknesses could become far easier to find and exploit when autonomous systems can probe networks continuously, adapt their approach and repeat attacks at a scale impossible for human operators.

Bradshaw warned that replacing vulnerable legacy technology across government and industry could require “enormous” expenditure.

That leaves governments facing an awkward collision between two technological eras.

Much of today’s critical digital infrastructure was built when cyber attacks ultimately depended on humans identifying targets, testing vulnerabilities and deciding what to do next.

AI agents increasingly remove those constraints.

A vulnerability that once required skilled attackers to discover and exploit manually could potentially be searched for across thousands of systems simultaneously.

Bradshaw also backed giving national-security agencies a role in independent evaluations of advanced AI models, bringing intelligence and security expertise directly into attempts to measure what frontier systems can actually do.

Australia is already debating what guardrails should govern increasingly capable AI systems while simultaneously adopting them inside government. Australian authorities are using Anthropic’s advanced Claude Mythos model, illustrating how quickly the same technology is becoming both a government capability and a national-security concern.

The debate around advanced AI frequently gravitates towards hypothetical superintelligence, loss of control and systems whose capabilities exceed those of humans across almost every domain.

Cyber security may present a considerably nearer problem.

Governments and businesses have accumulated decades of technology debt while assuming that attackers, however sophisticated, would continue operating broadly at human speed.

That assumption is disappearing.

The first major infrastructure problem created by autonomous AI may not require superintelligence at all.

It may simply require machine-speed attackers to meet systems still defended for the age of human hackers.