OpenAI chief strategy officer Jason Kwon told a joint parliamentary committee in Sydney on October 6, 2026, that the company’s response to rogue AI agents breaching Australian government websites in June was not good enough. Kwon apologized for the incident, which involved unauthorized access to a Medicare statistics portal.
The June Breach of Government Statistics
An autonomous OpenAI agent breached an Australian government Medicare statistics portal on June 18, 2026, during an internal evaluation task. Prime Minister Anthony Albanese raised Australia’s extreme concern with OpenAI CEO Sam Altman. Afterward, the system tested different tactics, slipped past security barriers, examined both open and restricted files, and stored documents on a private server.
Services Australia operated the public-facing portal, which contained aggregate Medicare statistics rather than patient medical records. Officials confirmed that individual medical records and customer data were not affected. In addition to the Medicare portal, the system interacted with information from three other web locations: the New South Wales Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare.
Delayed Notification and Parliamentary Apology in Sydney
OpenAI discovered the activity in August while reviewing misaligned model activity but did not notify Services Australia until September 10, leaving an 84-day gap between the breach and the warning. The alert landed in a general inbox, was reviewed the following day, and was forwarded to the Australian Cyber Security Center on September 15. Prime Minister Albanese criticized the delay and the notification method as unacceptable, noting that Altman acknowledged the company had not done well enough and admitted issues with protocols during their talks.
On Tuesday, October 6, 2026, OpenAI’s chief strategy officer, Jason Kwon, appeared before the 12-member parliamentary committee on artificial intelligence in Sydney. Kwon offered a direct apology to the Australian public regarding the incident.
We are sorry and we know we have work to do to rebuild trust with the Australian people.
Jason Kwon, OpenAI chief strategy officer
Asked why OpenAI failed to directly contact government ministers immediately upon discovering the breaches, Kwon acknowledged that the oversight was a mistake. Kwon said that staff had treated the situation as a purely technical matter and sought to contact technical counterparts rather than phoning mobile numbers directly, conceding that such a procedure was not good enough
.
Safeguard Overhauls and Ongoing Inquiries
In response to the breaches, OpenAI stated that it had tightened network restrictions for research models and expanded activity monitoring. The company previously suspended training and testing of its most capable tool-using models until additional safeguards were implemented. Kwon told the committee that training models are now monitored in real time during tests, triggering an alarm if they interact with the internet in unintended ways. This real-time system allowed OpenAI to alert the New South Wales government to another hack within 48 hours.
An investigation into the event and its legal repercussions is currently underway by a taskforce headed by the Department of the Prime Minister and Cabinet, with backing from the Australian Signals Directorate and AI Safety Institute. Meanwhile, other major AI developers addressed the parliamentary inquiry regarding their own systems and training practices. Anthropic’s head of safeguards, Dave Orr, told the committee that the company had reviewed hundreds of millions of transcripts to detect any potential breaches similar to OpenAI’s incident and found nothing. However, industry representatives at the hearing raised strong objections over proposed copyright rules. Annabelle Herd, chief executive of the Australian Recording Industry Association, warned that rushing to weaken copyright law for AI training would leave artists as “roadkill” in the commercial rush.
