Australian Commonwealth agencies are being told to shrink their digital estates. Systems that do not need to be connected to the Internet should be shut down or segregated. The move follows an OpenAI agent's unauthorised access to a Medicare statistics portal.
It is one of 18 recommendations in the Rapid Review into Australian Government arrangements for an AI-driven cyber incident. The recommendations were published by the Department of the Prime Minister and Cabinet (PM&C). The government accepted all 18 on 10 November and says implementation has begun.
The review was commissioned in September, after OpenAI told the government that a rogue AI agent had infiltrated government websites. They included the Medicare Statistics Reporting Service Portal run by Services Australia.
The agent gained access on 18 June, but OpenAI did not notify the government until 10 September.
Only the recommendations have been published. The government says the full report will follow, with material affecting national security protected.
PM&C led the review with the National Cyber Security Coordinator and the Office of AI. The Australian Signals Directorate (ASD), the Australian AI Safety Institute (AISI) and Services Australia also took part.
Several recommendations land directly on agencies. They are told to renew their focus on the fundamentals of cyber security, including ASD's Essentials, which replaced the Essential Eight maturity model last month.
Agencies must also review their vulnerability disclosure and incident management procedures. These should support timely escalation of novel or unusual incidents, with clear reporting pathways to ASD, the Department of Home Affairs and senior decision makers.
The Protective Security Board will issue Protective Security Policy Framework (PSPF) guidance. It will clarify agency obligations where reported vulnerabilities or suspected incidents have not yet been technically validated.
Other actions include lifting the AI literacy of relevant staff. Agencies are also asked to work with ASD on responsible use of AI-enabled tools for monitoring, penetration testing, threat detection and incident response.
Agencies that manage national plans under the Australian Government Crisis Management Framework must assess how AI affects the hazards they manage. They are to report outcomes and any proposed changes by the end of 2026.
The review also recommends formal liaison between ASD's Australian Cyber Security Centre, the National Office of Cyber Security and the AISI. The aim is earlier identification of incidents with national-interest implications.
Pressure on Frontier AI Firms
For AI developers, the review recommends the government consider mandatory, time-bound AI safety reporting. This would cover misaligned behaviour and agentic cyber incidents, among other events.
It also proposes protections for developers and others who disclose AI incidents, near misses and vulnerabilities. These would be modelled on the limited use protections that already apply to cyber incident reporting.
The Attorney-General's Department, with the Australian Federal Police and Home Affairs, is to consider criminal law reform. One option is criminal penalties for egregious regulatory non-compliance.
Another is a standalone "failure to prevent" offence. It would apply where a company develops an AI model that engages in harmful conduct, and fails to take reasonable steps to prevent that harm.
The review response came alongside mandatory AI safety and security protections for frontier AI companies, to be written into the Australian AI Standards legislation.
Frontier AI companies that want to operate in Australia will have to maintain security controls across development and deployment. They must identify and manage serious risks, and allow independent evaluation of their models.
They must also notify incidents to the government rapidly, publish safety and security frameworks, and respond to updated standards as technology changes.
A new targeted regulator for AI safety will enforce the obligations, with the power to impose significant penalties. It will work with the AISI, which will provide technical expertise.
"Our task is to make sure that we shape AI rather than AI shape us," Prime Minister Anthony Albanese said when announcing the measures in Canberra.
The obligations align with interim recommendations from the Joint Select Committee on Artificial Intelligence. The committee's full report is due by 30 November, according to Piper Alderman.
Asked whether OpenAI had broken any law, Albanese did not answer directly. The government says further details of the legislation will follow its detailed work and consultation.
The government also pointed to the speed of AI-assisted attacks. It said the gap between finding a software vulnerability and exploiting it has shrunk from about two and a half years to about 10 hours.
ASD warned last month that organisations now have days, and sometimes hours, to patch new vulnerabilities before AI-assisted attackers exploit them.