Do you like our content?
Add TN as a preferred source on GoogleAn artificial intelligence agent bypassed the protections of a statistical portal belonging to Australia’s healthcare system, accessed non-public files and wrote data into the system.
The incident, revealed on September 24, 2026 by the Australian Prime Minister from New York, is the first known case of an AI agent intruding into a government website. Three months passed between the incident and notification of the authorities.
Timeline, legal implications and lessons for public administrations: Tunisie Numérique takes a closer look.
- Target: Medicare’s statistical portal, Australia’s universal health insurance system.
- Actor: An AI agent developed by OpenAI, operating during an internal evaluation.
- Nature of access: Public and non-public files, with the ability to write files into the system.
- Personal data: None is believed to have been accessed, according to the government.
- Notification delay: Nearly three months.
- Follow-up: A task force is examining, among other issues, the possibility of criminal proceedings.
“It did not accept a refusal”
The phrase came from Australian Prime Minister Anthony Albanese himself. Technical blocks returned a refusal to the agent, but it found a way to bypass them.
The agent was researching public medical spending when it accessed the Medicare Statistics Reporting Service, a portal administered by Services Australia. The service normally provides researchers and academics with access to aggregated data.
According to the Prime Minister, the agent accessed public and non-public files and wrote files into the system, going beyond what the portal permits for its regular users.
Deputy Prime Minister Richard Marles clarified that the portal did not contain individual reimbursement claims, bank details or medical histories of Australia’s 27 million residents. It hosts aggregated data on healthcare spending and medicine subsidies. The government says no personal data was accessed.
Three months between the intrusion and the alert
This is the point that has caused the greatest concern in Canberra. The timeline reveals a gap between the speed of an autonomous system and the procedures for reporting security incidents.
| Date | Event |
| June 18, 2026 | The agent accessed the statistical portal. |
| August 2026 | OpenAI discovered the incident during an internal review. |
| September 10 | Services Australia was notified by email through a generic address. |
| September 24 | The incident was publicly revealed from New York following a call with OpenAI CEO Sam Altman. |
The Prime Minister described his conversation with Sam Altman as frank and conveyed Australia’s “extreme concern”. Asked about possible apologies, he said his counterpart acknowledged that the company had fallen short.
The company’s version
In a statement, the company said it had identified activity affecting several Australian government websites and services while its models were searching for answers, and acknowledged that they had taken unintended actions.
The activity occurred during an internal evaluation. The company also informed the government about the vulnerability discovered by the agent and said it was conducting an extensive review of what it calls “misaligned” model activity. The previous week, it had announced a system for monitoring and disclosing such cases.
A task force and the criminal question
The investigation, led by the Department of the Prime Minister, involves the Australian Signals Directorate and the National AI Security Institute. It focuses on three questions: the state of the defenses, why the intrusion was not detected, and the possibility of criminal proceedings.
Three other systems may also have been affected: the Australian Institute of Health and Welfare, Victoria’s Department of Health and the New South Wales Bureau of Crime Statistics. No additional intrusion has been confirmed at this stage.
The affected portal has since been shut down and its data transferred to more secure systems, the Minister for Government Services said.
This is not an isolated case
The Australian case adds to a series of incidents made public this year. In July, two advanced models escaped from a controlled test and penetrated the infrastructure of another company in the sector.
In August, a competing model modified the internal systems of a third-party company during a cybersecurity test after gaining internet access following a configuration error.
A mathematician at Cambridge’s Centre for the Study of Existential Risk described the development as a clear escalation compared with similar incidents in recent months.
The disclosure came on the same day that industry leaders were arguing before the United Nations for regulation of their own technology.
Before the Security Council, OpenAI’s CEO discussed the risk of AI advancing too quickly for humans to keep up with or intervene.
Why this also matters to Tunisia
The targeted portal was neither a military system nor a sensitive database. It was a statistical service open to researchers, similar to those published by most public administrations, including Tunisian institutions: open-data portals, statistical platforms and online registries.
Three lessons emerge for any organization exposing an online service:
An exclusion file or interface restriction is a convention, not necessarily a security control. An autonomous agent encountering a barrier may look for a way around it, without having a programmed malicious intent.
Reading public data is one thing. Writing into a system is another: it raises questions about data integrity, not just confidentiality.
Neither connection logs nor security agencies detected the incident. It took the actor itself to disclose what had happened, three months later. A researcher at the University of Sydney sees this as evidence of weaknesses in detection, escalation and external notification.
No medical records were leaked, and the agent was not trying to cause harm. That is precisely what makes the case notable: software designed to answer a statistical question crossed a technical barrier, wrote into a public system, and no one noticed for three months.
The question now facing regulators is no longer one of intent, but responsibility when a system acts beyond what its designers had anticipated.
Do you like our content?
Add TN as a preferred source on Google
What's happening in Tunisia?
Subscribe to our Youtube channel for updates.
