An OpenAI artificial intelligence agent accessed multiple Australian government systems / Alert sent by OpenAI to the authorities, also drafted by an AI agent

An artificial intelligence agent developed by OpenAI accessed Services Australia data and three other systems in June, and the company subsequently used AI to draft part of the email notifying the authorities. The information was revealed by The Guardian Australia.

The incident occurred on June 18, but OpenAI did not notify Services Australia until September 10. The company had learned of the issue in August, while the initial message was sent to an address checked only once a day. The email was partially drafted with the help of algorithms, including its wording and formatting, but was reviewed and sent by employees.

The notification stated that the agent exploited a vulnerability in the Medicare statistics service, without an account or password, and was able to read files and internal settings, list available files, and create a test file. OpenAI found no evidence that patient data had been accessed, information deleted, or permanent access established.

Chief Strategy Officer Jason Kwon acknowledged that the company’s response had been delayed. Assistant Minister Andrew Charlton called for stricter rules, saying that the agent had illegally breached a government system and that safety cannot be left solely to the market.

[/p] [p]The notification stated that the agent exploited a vulnerability in the Medicare statistics service, without an account or password, and was able to read files and internal settings, list available files, and create a test file. OpenAI found no evidence that patient data had been accessed, information deleted, or permanent access established.

Chief Strategy Officer Jason Kwon acknowledged that the company’s response had been delayed. Assistant Minister Andrew Charlton called for stricter rules, saying that the agent had illegally breached a government system and that safety cannot be left solely to the market.