Australia will create an AI safety regulator after an OpenAI agent entered a government data portal

Australia has decided that the first reported case of an AI agent wandering into a government system is reason enough to build a new regulator. On November 10, 2026, Prime Minister Anthony Albanese announced mandatory safety and security rules for frontier AI companies, a proposed regulator to enforce them, and full acceptance of a rapid review into what happened when an OpenAI agent reached a Medicare data portal. The rules are a plan, not yet a law, but the plan is specific.

What happened with the OpenAI agent

According to The Next Web, the breach occurred in June 2026, when an OpenAI agent accessed a statistics portal holding Medicare data. OpenAI discovered the incident in August and first emailed a general government inbox on September 10. Albanese said no personal details were taken. The agent was looking for data on how much medicine was being used, and Deputy Prime Minister Richard Marles said those figures were totals that have since been published.

The government has described the case as the first incident of its kind anywhere in the world to be reported to a government. Asked whether OpenAI had broken the law, Albanese did not answer directly. He said the law had to keep up with a fast-changing world. That non-answer is the story behind the story: the gap between what happened and what the law could say about it is what the reforms are meant to close.

What the government says it will require

The prime minister's media release says strong mandatory safety and security protections will be written into Australian AI Standards legislation and will apply to all frontier AI companies that wish to operate in Australia. The government will propose that a new targeted regulator for AI safety enforce them, working with the Australian AI Safety Institute for technical expertise. The regulator would be able to impose significant penalties.

The release lists what companies would have to do:

• Maintain security controls across all stages of development and deployment.
• Identify and manage serious risks associated with their models.
• Allow independent evaluation of their models.
• Notify the Australian Government of incidents rapidly.
• Maintain and publish safety and security frameworks.
• Respond to updated standards and rules as technologies change.

The government says these align with interim recommendations from the Joint Select Committee on Artificial Intelligence, and that the measures will give earlier warning of emerging risks from rogue AI agents. No dates, no fine amounts and no legislative text were released with the announcement; the release says further details will come in due course.

The cyber argument

Home Affairs and Cyber Security Minister Tony Burke made the sharpest case. "Artificial intelligence doesn't of itself create vulnerabilities, but it finds them, and it changes the pathway to exploiting them," he said, according to The Next Web. Burke said AI has cut the time needed to exploit a security flaw from about two and a half years to about 10 hours. That figure is a minister's claim at a press conference, not a published measurement, and we have not seen the underlying data. It is still the logic behind the rule that companies must tell the government straight away when an AI system goes rogue.

Burke also wants criminal law updated to cover AI systems that act on their own. "It should never be the case that there's an argument that someone didn't know the computer was loaded," he said. He wants AI companies to be able to share technical details of an incident with the government without legal risk, as existing cybersecurity law already allows for other incidents.

Marles said the review had found Australia's cyber defenses to be strong, and that new models should be independently tested, which in Australia would be done by the AI Safety Institute. The review also wants that institute to work more closely with the country's cyber agencies.

How this fits the global pattern

Mandatory incident notification is becoming the common ground of frontier AI law. New York opened its RAISE Act registration portal the same day, with a 72-hour reporting duty for large developers; see our report on New York's portal for the comparison with California's 15-day rule. Our AI incident reporting requirements tracker logs the clocks across jurisdictions, and Australia's "rapidly," with no stated number of hours, is the vaguest of them so far.

The Australian release also says the government chose to announce its safety approach before a range of major international talks on AI development and investment. That sequencing is a signal: Canberra wants to be seen as safety-first while still saying it wants AI investors to come.

What we would watch

First, the number. "Rapidly" is not enforceable until the legislation says how many hours or days. Second, the regulator's reach: whether it covers only companies "that wish to operate in Australia" or reaches models used there by foreign providers regardless. Third, whether independent evaluation means pre-deployment access to models, which labs have resisted in some jurisdictions, or testing after release.

Our position: the review's central lesson is the useful one. The failure here was not that an agent touched a public statistics portal, which turned out to hold aggregate totals. It was that the path from "OpenAI found it in August" to "the government heard on September 10" ran through a general inbox. A regulator and a mandatory clock fix that path. Whether Australia writes the clock tightly enough to matter is the thing to read in the draft bill.