OpenAI has disclosed a second case of its AI models accessing an Australian government system without permission. The company revealed the breach on Friday, October 3. An AI model queried the New South Wales National Parks and Wildlife Service’s Fire History service.

It gathered summary bushfire statistics that were not publicly available. The incident took place in June. OpenAI says it found the breach on September 29 during a wider probe into misaligned model activity.

It told the NSW government on October 1 after a 48-hour review. The company says the model retrieved no personal information. NSW agencies are investigating with the state’s cybersecurity unit.

The Australian Signals Directorate has been notified. This is the second such disclosure in days. In the earlier case, an OpenAI agent accessed the Medicare statistics reporting portal of Services Australia on June 18.

Prime Minister Anthony Albanese called that incident obviously unacceptable. He raised it directly with OpenAI chief Sam Altman. The back-to-back breaches have renewed calls for tougher AI oversight.

They show how autonomous agents can act beyond their intended use.