OpenAI AI agent bypassed restrictions on Australian government health website while searching for government data

OpenAI AI agent bypassed restrictions on Australian government health website while searching for government data

The incident highlights growing concerns around AI agents acting autonomously when they encounter restrictions, even as Australian authorities say there is currently no evidence of personal data being accessed.

Advertisement
    Share:
ऑस्ट्रेलियाई प्रधानमंत्री ने मुनाफाखोरी करने वालों पर एक्शन की बात कही है.OpenAI is facing scrutiny after an AI agent accessed non-public files on an Australian health portal.
Business Today Desk
  • Sep 24, 2026,
  • Updated Sep 24, 2026 12:00 PM IST

An OpenAI AI agent gained unauthorised access to an Australian government health website in June after attempting to get around restrictions while searching for information on medicines and health statistics. The case concerned the Medicare Statistics Reporting Service portal, which is managed by Services Australia, and has raised concerns as to what action AI agents might take when they come across obstacles while carrying out a task.

Advertisement

Must Read: ChatGPT can now handle work tasks by voice on your phone: Here’s what you can do

Prime Minister Anthony Albanese said the agent had accessed both public and non-public files after being prevented from entering a restricted area. Although at present there is no evidence that any personal information or patient records were accessed, Australian authorities are carrying out an investigation into the incident, including to find out whether any other government systems were affected.

OpenAI agent tried to bypass restrictions

Albanese stated that the incident took place in June as part of OpenAI's training exercises designed for the purpose of evaluating its AI models. The agent had been instructed to search the internet for information concerning Australian government expenditure on medicines and health statistics.

Advertisement

Must Read: Claude finds novel enzyme in DNA: Could AI be entering the gene-editing discovery race? 

When it came across a restricted area, the agent allegedly tried to find a way around the restriction and then accessed files that were not publicly available. OpenAI spotted this behaviour as part of its examination of what it referred to as "misaligned model activity", explaining that the models had been trying to obtain answers and statistics during an internal evaluation but had carried out actions which were not intended for them.

No evidence of patient data accessed

The incident is concerning, but authorities have so far found no evidence that patient records or personal information were accessed. The information involved aggregate health statistics and internal file names.

Advertisement

Albanese added that there was no evidence that the wider Services Australia network had been compromised.

Australian authorities investigating

The Australian Signals Directorate is helping out with a forensic investigation in order to find out precisely what took place and to determine whether any other government systems were affected.

Albanese stated that he had spoken to Sam Altman, the CEO of OpenAI, regarding the matter and had conveyed Australia's concerns. Having verified the information accessed, OpenAI informed Services Australia on September 10. Albanese criticised the delay in notifying the government, while OpenAI said that its review is still ongoing and that it is supplying technical information to assist with the investigation. 

For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine

An OpenAI AI agent gained unauthorised access to an Australian government health website in June after attempting to get around restrictions while searching for information on medicines and health statistics. The case concerned the Medicare Statistics Reporting Service portal, which is managed by Services Australia, and has raised concerns as to what action AI agents might take when they come across obstacles while carrying out a task.

Advertisement

Must Read: ChatGPT can now handle work tasks by voice on your phone: Here’s what you can do

Prime Minister Anthony Albanese said the agent had accessed both public and non-public files after being prevented from entering a restricted area. Although at present there is no evidence that any personal information or patient records were accessed, Australian authorities are carrying out an investigation into the incident, including to find out whether any other government systems were affected.

OpenAI agent tried to bypass restrictions

Albanese stated that the incident took place in June as part of OpenAI's training exercises designed for the purpose of evaluating its AI models. The agent had been instructed to search the internet for information concerning Australian government expenditure on medicines and health statistics.

Advertisement

Must Read: Claude finds novel enzyme in DNA: Could AI be entering the gene-editing discovery race? 

When it came across a restricted area, the agent allegedly tried to find a way around the restriction and then accessed files that were not publicly available. OpenAI spotted this behaviour as part of its examination of what it referred to as "misaligned model activity", explaining that the models had been trying to obtain answers and statistics during an internal evaluation but had carried out actions which were not intended for them.

No evidence of patient data accessed

The incident is concerning, but authorities have so far found no evidence that patient records or personal information were accessed. The information involved aggregate health statistics and internal file names.

Advertisement

Albanese added that there was no evidence that the wider Services Australia network had been compromised.

Australian authorities investigating

The Australian Signals Directorate is helping out with a forensic investigation in order to find out precisely what took place and to determine whether any other government systems were affected.

Albanese stated that he had spoken to Sam Altman, the CEO of OpenAI, regarding the matter and had conveyed Australia's concerns. Having verified the information accessed, OpenAI informed Services Australia on September 10. Albanese criticised the delay in notifying the government, while OpenAI said that its review is still ongoing and that it is supplying technical information to assist with the investigation. 

For Unparalleled coverage of India's Businesses and Economy – Subscribe to Business Today Magazine

Read more!
Advertisement