AI agents in Australia have gone from accidentally hacking your local gym to trying to access government websites without permission with the speed of a trashy science fiction novel, and it’s left some observers – and headline writers – worrying about all the wrong things.
While there is a certain level of justifiable alarm at the pace of AI development and the manner in which AI can accidentally compromise other systems, the real worry, contends SailPoint’s Identity Strategist, APAC, Gary Savarino, is how unprepared most modern networks are for this kind of intrusion.
"The Medicare portal incident is less about rogue AI and more about what happens when goal-seeking behaviour meets inadequate preventative and detective controls,” Savarino told Cyber Daily.
“This wasn't malicious intent on the part of the AI agent; it was machine-speed persistence. Had a human attempted the same route, the underlying security gaps might have produced the exact same result.
“The critical difference is that an AI agent operates at machine speed and scale. Think of it less like a malicious actor and more like an overly diligent teenager who hasn't been given strict boundaries. Give an agent a task and it will hunt down the quickest path to complete it. If the obvious route is blocked, it won't give up, it will iterate, probe and discover accessible, non-public files far faster than any human could, regardless of whether that outcome was intended.”
Savarino said the response from the likes of the National Cybersecurity Coordinator, the Australian Signals Directorate, and the Australian AI Safety Institute shows that AI agents are now becoming a serious policy priority. However, what the incident really revealed was the “systemic challenge” faced by both public and private sectors, and “that organisations still lack the baseline visibility and controls required for an agentic world”.
“Whether an identity is human or non-human, the core questions remain the same. What can this identity access, are proper boundaries in place, and do we have the access controls to catch anomalous activity in real time?” Savarino said.
“In sectors like healthcare and government, where citizen data is highly sensitive, relying on assumptions is no longer an option. Left ungoverned, AI agents will rapidly uncover misconfigurations, scavenge credentials and inherit excessive permissions. Legal reform and new standards will take time, but the need to enforce least privilege and continuous monitoring exists right now.
“To safely embrace agentic AI, organisations must implement the identity security tools needed to establish clear boundaries, detect when an agent starts to drift from approved policies, and swiftly revoke access before privacy is compromised."
Want to see more stories from trusted news sources?Make Cyber Daily a preferred news source on Google.
