Scrutiny over AI agents operating beyond their intended boundaries is growing across the region, after an OpenAI agent accessed an Australian government health portal through what appears to have been a misconfigured website. The incident, and the delay before affected parties were notified, has reopened a question security teams have been circling for months: who is responsible when an autonomous system does something it wasn’t supposed to do?

Singapore has joined international calls for frontier AI systems to remain under human oversight, part of a broader push among regional governments to keep governance frameworks in step with how quickly autonomous systems are being deployed. On the industry side, Nvidia has introduced a dedicated agent-safety platform, a sign that vendors are starting to build safeguards specifically aimed at agents that stray outside their assigned tasks.

Old weaknesses, found faster

Dan Schiappa, President of Technology and Services at Arctic Wolf, said the Australian incident may say less about AI breaking through security controls than about AI finding the cracks that were already there. “Misconfigurations have long been a thorn in the side of organizations when it comes to unintended data exposure, and AI agents are exceptionally good at finding those soft spots in defenses,” he said. “AI agents operate at a speed and scale that uncover configuration issues faster than any human. On top of that, they do not need malicious intent to create legitimate security concerns.”

Schiappa said the responsibility now falls on security teams to find those configuration issues first, limit access to what is strictly necessary, and account for every endpoint, including the ones no human ever interacts with directly.

Accountability can’t disappear because the actor is software

Adam Marre, Senior Vice President and Chief Information Security Officer at Arctic Wolf, argued that growing autonomy doesn’t let AI vendors off the hook for how their systems behave once deployed. “As AI systems become more autonomous, accountability cannot disappear because the actor is software,” he said. “If a human employee accessed systems without authorization or bypassed established rules, there would be consequences and a clear expectation that the organization involved would address the problem. The same standard should apply when AI systems are responsible.”

Marre added that government systems should not become a testing ground for the unintended consequences of autonomous technology. “AI companies have a responsibility to ensure their products operate safely in the real world and to move quickly when failures occur,” he said. “Until the industry treats these failures with the same seriousness it would if a human caused them, we should expect to see more incidents like this.”