An OpenAI agent accessed Australian government health data in June 2026. The government was not told until September. That gap – nearly three months between breach and disclosure – is not just a political problem. It is a cyber insurance problem, and one that brokers with public sector and health sector clients need to examine now.
Prime Minister Anthony Albanese confirmed on September 24 that an OpenAI agent had accessed a Medicare data portal, reaching both public and non-public files, though no personal Medicare details appeared to have been compromised, according to ABC News.
OpenAI did not notify the federal government until September. Subsequent reporting from the International Association of Privacy Professionals (IAPP) confirmed the agent accessed the portal on June 18 and that OpenAI’s notification arrived via a generic public inbox – 84 days after the incident.
On the same day Albanese made his announcement, the NSW Bureau of Crime Statistics and Research (BOCSAR) was notified by the Australian Signals Directorate (ASD) that a dataset underpinning a publicly accessible crime mapping tool had been flagged by OpenAI as potentially vulnerable. BOCSAR said there was no evidence the dataset had been accessed or that a breach had occurred.
In a statement, the agency said: “While the dataset does not contain personal information and any risk associated with exposure is considered low, BOCSAR is reviewing the concerns raised and working to strengthen its systems and safeguards.”
Albanese told a press conference in New York that BOCSAR, the Australian Institute of Health and Welfare, and the Victorian Department of Health “may (also) be impacted.”
Read next: Medicare AI breach tests how cyber wordings define unauthorised access
Under Australia’s Notifiable Data Breaches (NDB) scheme, regulated entities must notify the Office of the Australian Information Commissioner (OAIC) and affected individuals as soon as practicable after becoming aware of an eligible data breach, according to the OAIC. NSW public sector agencies operate under a parallel mandatory scheme introduced in November 2023 that mirrors the federal framework.
The coverage question the Medicare incident raises is specific: most cyber policies measure the notification window from when the insured “knew or ought to have known” – not when a third-party vendor chose to disclose. Where those two dates are 84 days apart, the gap can determine whether a claim is valid.
NSW Premier Chris Minns, who said he had spoken with Albanese about the matter, identified the AI agent’s conduct as the central concern. “As I understand it the AI agent was told not to access these parts of the website, this information, and it did so anyway. Now that’s a big concern for us in New South Wales. We need to and will examine its impact both on technology and the vulnerabilities in our confidential information or information that is not public-facing,” Minns said.
That description – a system acting outside its own instructions – sits against standard cyber policy language that was written for human-initiated intrusions. Kieran Doyle, partner and head of cyber, data, and technology at Wotton Kearney, noted in the context of this year’s 5-Star Cyber Report that non-ransomware incidents, including privacy breaches, represent a category of risk that has “always been there but perhaps went undetected until now.”
On how the market is treating AI risk, Serene Davis, QBE’s global head of cyber, said: “AI is treated as a risk amplifier, not a fundamentally new cyber risk.” Greg Eskins, global cyber product leader at Marsh, has stated that underwriters recognise “it’s important to continue to offer a product that responds to these types of events.” Both comments were made in the context of how insurers are responding to agentic AI incidents more broadly.
That is the majority market position – but policy language varies between insurers, and wordings have not yet been tested against this precise fact pattern. Landers & Rogers noted in March 2025 that explicit AI exclusions in Australian liability policies remain limited but that the position is shifting as AI-related claims begin to test existing wordings.
NSW Greens industry spokeswoman Abigail Boyd called for a broader audit of all state government systems, pointing to the volume of data sitting in public-facing portals. “BOCSAR is not the only vulnerable source of potentially sensitive information. The NSW government hosts highly sensitive information through public-facing websites like the HealthStatsNSW website, as well as the Data.NSW website which is an open data portal hosting 17,164 NSW public sector datasets,” Boyd said.
For underwriters assessing public entity risks, that figure gives an indication of the exposure surface. The OAIC recorded 1,205 data breach notifications across Australia in 2025 – the highest annual total since mandatory reporting began in 2018, up 8% on 2024 – with health service providers accounting for 19% of all notifications and Australian government agencies contributing a further 118.
Read next: What the Gemini and Claude hacking incidents mean for cyber insurers
The federal government confirmed that forensic investigation findings will feed into national AI standards in development, according to ABC News. Proposed regulations may include mandatory breach reporting for AI companies whose products are involved in security incidents. AI-specific legislation is being targeted from 2027.
OpenAI CEO Sam Altman and Anthropic CEO Dario Amodei addressed the UN Security Council on the same day the breach was made public, both calling for international cooperation on AI governance.
Minns framed the significance of those calls. “These are the people who will most specifically financially benefit from the expansion of AI. If they’re saying it should slow down, I think that everybody else should heed the warning and look very closely at what this very powerful, potentially damaging technology can do with public information,” Minns said.
For brokers with public sector, health, or government-adjacent accounts, the question is not whether the regulatory environment shifts – it is whether existing policy wordings would respond to the class of incident that just occurred, and whether that has been confirmed ahead of renewal.
