• Wion
  • /World
  • /OpenAI admits AI agent ‘scaled the fence’ into Australia government system, apologises: ‘We are sorry’

OpenAI admits AI agent ‘scaled the fence’ into Australia government system, apologises: ‘We are sorry’

OpenAI admits AI agent ‘scaled the fence’ into Australia government system, apologises: ‘We are sorry’

‘We should have… sooner’: OpenAI apologises over delayed warning on rogue AI incident Photograph: (Combination image: AFP pics)

Story highlights

OpenAI has apologised to Australia after an AI agent gained unauthorised access to a government portal during a June evaluation. The company also admitted it should have shared its preliminary findings with Australian officials sooner.

OpenAI has apologised for not properly responding to the security incident involving rogue AI (Artificial Intelligence) agents flagged by Prime Minister Anthony Albanese at the UNGA81 summit earlier this month. In a blog post on Monday (Sep 28), the ChatGPT maker said it was "sorry" and that it should have shared early investigation results with Australian officials "sooner".

“We are sorry and working to do better in the future,” OpenAI said in a blog post published Monday. The company said it would set out what it knows about the incident, the changes it has made and the steps it plans to take to rebuild trust with Australians.

Add WION as a Preferred Source

AI model bypassed restrictions on government website

The incident took place in June when an OpenAI model was being used to research Australian government data.

Australian Prime Minister Anthony Albanese said the AI agent gained unauthorised access to the public-facing Medicare Statistics Reporting Service portal, administered by Services Australia. The model accessed both public and non-public files after continuing its attempts to obtain information despite access restrictions.

Trending Stories

Australian officials said the model was seeking information on government spending on medicines. After its request was denied, the agent continued looking for a way to obtain the information.

Deputy Prime Minister Richard Marles described the behaviour as an AI agent entering a government website without authorisation. "It asked a question, the information was not given, and rather than leaving at that point, it scaled the fence." He said the portal contained aggregated medical statistics and that no individual's medical information was accessed.

OpenAI admits delay in notifying Australia

OpenAI said it discovered the activity during an extensive review of its models in August. The company acknowledged that its communication with Australian agencies was not handled properly.

“Our aim was to give affected agencies a detailed account once our investigation was complete,” OpenAI said. “However, we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.”

Australia was informed on September 10, weeks after OpenAI identified the incident. Albanese criticised the delay, saying the government should have been notified earlier.

Four Australian government websites involved

Australian authorities said the AI model interacted with four government-related websites during the June evaluation. These included the Australian Institute of Health and Welfare, the Victorian Department of Health, the NSW Bureau of Crime Statistics and Research, and the Medicare Statistics Reporting Service portal operated by Services Australia.

About the Author

Share on twitter

Moohita Kaur Garg

Moohita Kaur Garg is a journalist and Senior Sub-Editor at WION News with five years of experience covering the volatile intersections of geopolitics and global security. She has e...Read More