AI safety regulator to be created in Australia after OpenAI breach

Akhil Ponnada
Akhil Ponnada is the founder and editor of FrontWord. He covers artificial intelligence, chips and infrastructure, consumer technology, cybersecurity and the business of tech.
6 Min Read
Image: The Economic Times

Australia will establish a dedicated AI safety regulator and introduce mandatory incident reporting with potential criminal penalties for developers, Prime Minister Anthony Albanese announced in Canberra on Tuesday, following an unauthorised security breach of a government system by an autonomous OpenAI research agent.

The announcement follows a rapid government review into an incident where an OpenAI agent broke into a federal health portal, as Guardian Technology reported. “We accept all of the recommendations, and we’ve commenced work to implement them,” Albanese said of the 18 recommendations.

Powers of the planned watchdog

The new watchdog will oversee security controls throughout the development and deployment of artificial intelligence systems, as reported by Bloomberg. Albanese said the body will be empowered to conduct independent evaluations of frontier models, identify serious risks, and enforce “significant penalties” against non-compliant technology firms.

Under the proposals accepted by the cabinet, the government plans to include mandatory reporting obligations in draft legislation expected in 2027, as Reuters reported via Livemint. If enacted, artificial intelligence developers would face strict time limits to notify authorities whenever an autonomous agent displays misaligned behaviour or causes a cyber incident.

The review also suggested establishing protections for whistleblowers and developers who report near misses and vulnerabilities. Defence Minister Richard Marles said independent risk assessments are essential because artificial intelligence will continue to expand across society, adding that companies must open their systems to external scrutiny.

“Our task is to make sure that we shape AI, rather than AI shape us,” Albanese said, as quoted by Economic Times Tech. The prime minister stated that public social licence is critical for the country as it seeks to capitalise on the technology.

How the OpenAI agent breached the portal

The incident occurred on 18 June 2026, when an OpenAI research team used an internal model to gather information on Australian public medicine expenditure. According to Unite.AI, the agent was repeatedly blocked by system controls, but it generated alternative routes to bypass those safeguards and accessed non-public files on a Services Australia website.

The standalone portal hosted aggregate statistics covering Medicare and the Pharmaceutical Benefits Scheme. Government ministers confirmed that the breached section contained statistical data rather than claims, individual records or payment information, and OpenAI said an internal review found no evidence patient data was compromised.

OpenAI said the intrusion was unintentional. Days after the breach became public in September, the company delayed the rollout of a new artificial intelligence model, citing safety concerns.

A three-month delay in notification

The breach exposed vulnerabilities in how frontier technology labs detect and disclose autonomous system failures. OpenAI stated that it only identified the unauthorised activity in August while reviewing misaligned model logs, approximately two months after the agent breached the government website.

On 10 September 2026, OpenAI sent an alert to a general vulnerability-disclosure email address at Services Australia that staff checked only once a day. Services Australia validated the warning and alerted the Australian Signals Directorate five days later, according to the notification chain described by ministers.

The federal government revealed the breach on 24 September. The BBC reported that Albanese spoke directly with OpenAI chief executive Sam Altman to convey Australia’s deep concern, with Altman conceding there were protocol failures at his company.

In response to the delayed escalation, the Protective Security Board will issue guidance requiring government departments to alert senior decision-makers and cybersecurity officials before technical verification is finished. Minister for Government Services Katy Gallagher said the affected health statistics portal has been decommissioned permanently, with remaining public datasets shifted to data.gov.au.

Proposed criminal penalties and next steps

The government review instructed the Attorney-General’s Department, consulting the Australian Federal Police and the Department of Home Affairs, to explore criminal sanctions for serious corporate misconduct.

One proposal under review is a standalone failure-to-prevent offence. The statute would penalise technology companies if a model they build causes harm and the firm failed to implement reasonable preventative measures.

Tensions between Canberra and US artificial intelligence developers have intensified after Australia declined requests from OpenAI and rival Anthropic for training exemptions from domestic copyright laws. Both companies told an Australian parliamentary inquiry in October that they favoured mandatory breach reporting, acknowledging notifications currently rely on company discretion.

Before legislation is introduced in parliament next year, government agencies overseeing national crisis plans must evaluate AI hazards and submit their response arrangements to Crisis Arrangements Committee chairs by the end of 2026.

Follow:
Akhil Ponnada is the founder and editor of FrontWord. He covers artificial intelligence, chips and infrastructure, consumer technology, cybersecurity and the business of tech.