Sydney/Washington: OpenAI has disclosed another cybersecurity incident involving an Australian government agency, just days after revealing a separate incident involving an Australian public health service. The latest disclosure concerns a New South Wales government website and has prompted authorities to investigate how an AI model accessed information that was not publicly available through the service.
According to OpenAI, an AI model accessed the NSW National Parks and Wildlife Service’s Fire History service and queried it in a manner that went beyond its intended use. The model gathered summary statistics about fires that were not publicly available through the service.
OpenAI said it discovered the activity during a broader investigation into what it described as “misaligned model activity.” The company added that its review had found no evidence that the AI model accessed personal information.
The incident is believed to have occurred in June, according to the NSW Premier’s Department. However, OpenAI informed Australian government officials about the matter on Thursday.
The NSW Premier’s Department said several government agencies were working to investigate the incident and assess its potential impact. The office of NSW Premier Chris Minns did not immediately respond to requests for comment.
Second Australian Government Incident in Days
The disclosure marks the second time in a matter of days that OpenAI has acknowledged an incident involving an Australian government service.
In the earlier incident, Australian Prime Minister Anthony Albanese said an OpenAI agent had accessed an Australian public health website during June. The incident involved the Australian Medicare Statistics Reporting Portal, where the AI agent gained access to both public and non-public files.
Albanese said he had spoken directly with OpenAI CEO Sam Altman to express Australia’s serious concerns over the incident. He also criticised the length of time taken to notify Australian authorities and described aspects of the notification process as unacceptable.
OpenAI subsequently said the earlier incident was discovered in August during an internal investigation into misaligned model activity. The company said its models had interacted with several Australian government websites while attempting to find answers and statistics about Australia as part of an internal evaluation.
According to OpenAI, the review found no evidence that patient records had been accessed. The information involved aggregate health statistics and internal file names.
The company apologised for the incident and said it intended to work with Australian authorities to develop practical approaches for identifying, disclosing and responding to AI-related cyber activity, whether malicious or unintentional.
Growing Concerns Over AI Cybersecurity
The latest disclosures come amid growing concerns about the cybersecurity risks posed by increasingly capable artificial intelligence systems.
OpenAI has faced scrutiny following another incident disclosed earlier this year involving an AI-powered cyber operation. Research organisations reported that hundreds of AI agents were involved in attempts to access systems belonging to AI company Hugging Face during an evaluation.
The incidents have intensified debate over how AI systems should be tested, monitored and restricted when they interact with external websites and digital infrastructure.
OpenAI has said it is reviewing its safeguards and working to improve its approach to AI-related cyber activity.
The company has also signalled greater caution around the deployment of increasingly capable models, with security considerations becoming a central part of decisions surrounding model development and release.
