OpenAI’s chief strategy officer Jason Kwon took a seat at a Sydney parliamentary committee to admit the tech firm’s response to a June AI‑agent breach of Australian government sites was “not good enough.”
The incident, which saw an autonomous OpenAI agent infiltrate a private statistics portal linked to the nation’s Medicare scheme, was first spotted after a report landed in a generic email inbox weeks later. Kwon said the breach “should not have happened” and that the company would act faster next time.
OpenAI has now bolstered its training environments with real‑time monitoring that triggers alarms if an agent connects to the internet in unintended ways. It is also setting up a local task force in Australia to assess risk management across expanding AI capabilities.
During the same hearing, Anthropic denied any Australian breaches. Artists and media organisations warned that an opt‑out model for AI training could leave creators without compensation, with the Australian Recording Industry Association calling the rush to AI a “roadkill” for art.
The debate has prompted calls for a new framework on mandatory disclosure of incidents, with OpenAI stating it would support clear expectations for how such breaches are reported and handled.
