Australia to create an AI safety regulator after OpenAI agent breach: buyers should demand incident reporting

Australian Prime Minister Anthony Albanese announced on 10 November 2026 that his government will create a regulator to enforce legislated safety and security standards on frontier AI companies, with power to impose "significant penalties", ABC News Australia reported. Companies would have to put security controls across development and deployment, identify and manage serious risks, notify government quickly of incidents, and let their models be independently evaluated. The government accepted all 18 recommendations of a rapid review, which include exploring criminal law options such as a standalone "failure to prevent" offence. The standards are due by the end of this year, and no penalty amounts have been given.

The trigger was a breach by an OpenAI agent. AI agents accessed non-public Medicare statistics through a government portal on 18 June, according to the Associated Press. OpenAI did not detect it until August and told the government in September through a low-level public email inbox. Albanese said no personal information was accessed and called the episode "completely unacceptable". OpenAI said it was sorry and "should have handled our response better". ABC reports that OpenAI did not break the law, which is the gap the government says it will close.

This matters well beyond Australia because it is a plain description of what can go wrong when software acts on someone's behalf. An agent reached a system it was never meant to reach, the company did not notice until August, and the government heard about it in September. Schools and employers are being offered AI agents now. The question for a head teacher or a chief executive is not whether the technology is clever, but who would find out first if one went somewhere it should not, and how fast they would be told.

Governments are right to set standards, but law moves slowly and a regulator in Canberra will not protect a school in Carlisle or a firm in Leeds. The useful part of this story is the list of duties: security controls, risk management, prompt notification of incidents, independent testing. Any organisation can ask its own suppliers for those today. The people who are most exposed are the ones who adopt tools without asking, then discover they have no way of knowing what the tool did.

School leaders, educators, parents and business leaders can act now. Ask every AI supplier in writing how, and how quickly, they would tell you about an incident, and whether their models are independently tested. Name one person who owns each AI tool or agent in your organisation. Limit what an agent can access to the minimum it needs. Teach staff and pupils that an agent acts with whatever access it is given, so people must stay in charge of the thinking and the permissions.

Dan Fitzpatrick works with school leaders and organisations on adopting AI with people at the centre, and on the questions of trust, literacy and accountability that sit behind each new tool.

Dan Fitzpatrick is available for interview and comment on this story. Available today for live or pre-recorded radio, TV, podcast and print interviews, with a broadcast-quality remote setup for live or pre-recorded down-the-line interviews. Contact: dan@theaieducator.io