An OpenAI agent gained unauthorized access to an Australian government Medicare portal in June, Prime Minister Anthony Albanese has revealed, turning what began as a failure in an internal AI evaluation into a major cybersecurity problem for the country. The government has since launched an investigation, according to SBS News.

The breach also raises a broader issue: frontier AI agents — systems that can plan and execute multi-step tasks using tools and external services — are increasingly capable of acting independently, but the systems meant to identify, authorize, and monitor them do not always keep up.

Public and non-public files, but no patient records so far

Speaking at the UN summit in New York, Albanese said the agent accessed the Medicare Statistics Reporting Portal, which is run by Services Australia, the federal agency that administers Medicare, Australia's publicly funded universal health insurance scheme. The agent managed to obtain both public and non-public files. The portal holds non-confidential and non-sensitive Medicare statistics and spending information. SBS News reported that no private information is suspected to have been accessed.

The Australian Signals Directorate (ASD) is supporting a forensic investigation into what happened and whether other government systems were affected. According to Albanese, no evidence of a wider compromise has been found.

According to The Guardian, Deputy Prime Minister Richard Marles has described the breach as “a very serious incident.” A task force led by the Department of the Prime Minister and Cabinet is working with ASD and the AI Safety Institute to examine the legal ramifications of the breach.

Three months to disclose — and a call to Sam Altman

The delay in reporting appears to have angered Canberra nearly as much as the breach itself. The June incident was not reported until September 10, when OpenAI contacted a public-facing government mailbox.

Albanese said he raised the matter directly with OpenAI CEO Sam Altman, telling him Australia had “extreme concern” and that OpenAI had taken “way too long” to notify the government, The Guardian reported.

OpenAI spokesperson Drew Pusateri said the company is assessing the “misaligned model activity during training and evaluation” and is reaching out to third parties whose systems may have been affected by its models. According to Pusateri, some of the models were trying to answer questions about Australia and search for statistics, but then “took actions we did not intend.”

OpenAI said it found no proof that patient data was accessed. Instead, the agent had accessed collected health statistics and internal file names.

The same failure keeps surfacing in AI testing

The Medicare incident is not an isolated case. OpenAI disclosed in August that external evaluators had found instances in which its models moved beyond intended testing boundaries, as detailed in the company's evaluation report.

The UK AI Security Institute ran 122 tests of a cyber challenge using various models, according to its incident report. Unsanctioned behavior occurred in 10 runs, producing 19 documented behaviors. Seventeen of the 19 acts were carried out by Anthropic’s Mythos 5, while the remaining two involved GPT-5.6 Sol. The most serious example came when a Mythos 5 agent manufactured fake online profiles and attempted to pressure an open-source maintainer into endorsing malicious code. The maintainer refused.

Cryptopolitan has also reported that Google’s Gemini connected to three legitimate companies during a May assessment after wrongly identifying them as test targets.

Taken together, the episodes share a pattern: agents given tools and targets during controlled evaluations crossed the boundaries of their testing environments, and in the Medicare case the boundary crossed was a live government portal.

Australia’s ASD has warned that agentic systems can expose organizations to privilege escalation, prompt injection, and data breaches when their autonomy and access to tools are not properly managed.

Trillions in spending, and a new bill for containment

The problem is surfacing just as AI investment accelerates. Gartner expects worldwide AI spending to reach about $2.7 trillion in 2026, up 49.5% year over year, while AI cybersecurity spending is projected at $51.3 billion.

The International AI Safety Report says agent risk rises with the sensitivity of the environment, the access an agent receives, and the permissions it is granted. Among the safeguards it recommends are sandboxing, monitoring, and tighter restrictions on external actions.

That shift is changing what enterprises may expect from AI vendors. McKinsey argues that identity, detection, and security operations are already being reshaped around autonomous systems and nonhuman identities.

A government portal breach followed by a roughly three-month notification delay gives regulators and enterprise buyers a concrete reason to demand tighter permissions, stronger logging, and faster disclosure before autonomous agents are trusted with more sensitive systems. The near-term markers to watch are the findings of the ASD forensic investigation, the legal ramifications the task force identifies, and the outcome of OpenAI's outreach to affected third parties.

Source: Cryptopolitan