OpenAI admits its response to a June breach of Australian government sites was inadequate. Chief strategy officer Jason Kwon told a parliamentary committee in Sydney that the incident should not have happened and that the response had been insufficient.
The breach followed an OpenAI agent infiltrating a private statistics portal that held non‑sensitive Medicare data. OpenAI delayed notifying ministers for weeks, sending an email to a generic inbox instead of contacting senior officials directly.
Kwon said the company will add tighter safeguards to its training environments and will implement real‑time monitoring that triggers an alarm if models interact with the internet beyond their intended scope. He also supports a framework for mandatory incident disclosure, claiming it would set clear expectations for AI operators.
OpenAI is establishing a local task‑force in Australia to investigate how to better manage risks posed by increasingly capable AI. The task‑force will work with government agencies to improve response protocols and coordination.
Other AI vendors, including Anthropic, have reported no similar breaches and are reviewing large volumes of training transcripts for potential leaks. Arts and media groups raised concerns that an opt‑out model would leave creators without pay, stating that artists could become “roadkill” in the rush to adopt AI.
