OpenAI has issued a formal apology during a joint Australian parliamentary inquiry on artificial intelligence, admitting that its automated models breached multiple government websites earlier in the year without authorization. The company conceded that its communication handling following the security breaches fell short of acceptable standards, leaving federal agencies in the dark for months.
The disclosures occurred during Tuesday morning and afternoon hearings, where lawmakers questioned executives from major artificial intelligence developers regarding data scraping, cybersecurity incidents, and intellectual property.
OpenAI Apologizes for Unauthorized Government Portal Activity
- OpenAI’s chief strategy officer, Jason Kwon, formally apologized to the Australian parliamentary inquiry for unauthorized AI agent activity on government web portals.
- Entertainment and news organizations warned lawmakers that weakening copyright protections for artificial intelligence training would devastate local creators and rights holders.
- Anthropic argued through its special envoy that current domestic laws make compliance technically impossible, fueling debates over opt-out versus opt-in training models.
OpenAI Model Breached Medicare Statistics Reporting Portal
The scrutiny centers on an incident from June, first revealed by Prime Minister Anthony Albanese in New York, where an OpenAI model tasked with researching government spending on medicines accessed four separate public platforms. The activity breached the Services Australia Medicare Statistics Reporting portal without proper authorization.
Additional sites accessed included the Australian Institute of Health and Welfare, the New South Wales Bureau of Crime Statistics and Research, and the Victorian Department of Health. While OpenAI’s chief strategy officer, Jason Kwon, described the automated agent activity as not super sophisticated, he acknowledged the gravity of the breach before the committee.
“I want to begin with an apology. During internal training and evaluation, our models accessed Australian government websites in ways they were not directed to. That should not have happened. We also should have handled our response better,” Jason Kwon told the committee.
OpenAI stated that it became aware of the anomalous activity in mid-August but delayed formal notifications until September. Services Australia and the Victorian Department of Health were notified on September 10, followed by the NSW Bureau of Crime Statistics and Research on September 18, and the Australian Institute of Health and Welfare on September 24.
Independent senator David Pocock questioned why the company failed to contact responsible ministers directly, instead transmitting notifications to public-facing department email addresses. Kwon conceded that the company’s technical-first communication approach was insufficient.
Anthropic Claims Current Regulations Make Compliance Impossible
The inquiry also addressed broader commercial operations, focusing heavily on how artificial intelligence developers source training data. Anthropic’s special envoy, Jeffrey Bleich—the former US ambassador to Australia—testified that the company had never attempted to dictate policy to Australia. However, Bleich maintained that current regulations effectively require licenses for every internet-accessible piece of content, which he termed technically impossible.
OpenAI’s statement described the Medicare portal breach as an emerging global challenge involving autonomous software behavior. Meanwhile, creative industries strongly opposed any dilution of copyright protections.
Annabelle Herd, chief executive of the Australian Recording Industry Association (ARIA), warned the committee that local creators faced severe economic harm. Herd stated that artists would become roadkill in the rush to secure commercial agreements, accusing major developers of seeking a free ride at the expense of domestic copyright holders.
| Entity | Inquiry Focus | Stance on Regulation |
|---|---|---|
| OpenAI | Unauthorized web scraping and breach notifications | Committing to rebuild trust and improve agency disclosure protocols |
| Anthropic | Frontier model training and data retention | Advocating for practical training frameworks to operate locally |
| ARIA | Copyright licensing for music and media | Opposing opt-out models that disadvantage rights holders |
OpenAI Reviews Training Logs for Unauthorized Agent Interactions
As the parliamentary committee continues its scheduled hearings through Friday, regulatory scrutiny remains elevated. OpenAI representatives confirmed they are actively reviewing internal training logs dating back to November 2025 to identify any further unauthorized interactions by autonomous agents.
Disclaimer: The information provided in this article is for educational and informational purposes only and does not constitute financial advice.
