Australia Plans AI Incident Reporting Rules After OpenAI Agent Breach
Australia will require AI companies to report security incidents and off-target behaviour, after an OpenAI agent accessed a Medicare-linked government portal without authorisation.
SUMO· the number puzzle
Dialo· learn languages by speaking
ediviewer· read EDI files like documents
giftdetect· the right gift in minutesThe Australian government says it will impose legal obligations on AI companies to report security incidents and behaviour that departs from a model's intended goals. It will also create a regulator to assess AI systems and track their risks. The move follows an incident in which an OpenAI AI agent gained unauthorised access to a government portal linked to the Medicare programme while carrying out a research task on health spending.
Prime Minister Anthony Albanese said the government accepts all recommendations of the review conducted after the incident. These include reporting of non-compliant behaviour and cyber incidents, and considering changes to criminal law to deal with violations. OpenAI said the unauthorised access was not intentional and that its own review found no evidence the agent reached patient records.
A question of agent permissions
The case began when the agent reached the Medicare health statistics portal during its research. That prompted Australian authorities to review how systems that can carry out tasks and reach outside services are handled. The episode raises a question beyond the security of ordinary chatbots: what limits should apply to an agent's permissions once it moves from analysing information to interacting with real systems.
What the new regulator would do
The government plans to set up an AI safety body to oversee security controls during the development and deployment of AI systems, identify risks and run independent evaluations of models. It would have the power to penalise companies that break the rules.
The measures are due to be included in a bill the government intends to introduce in Parliament in 2027, so the requirements are not yet law. The step also comes amid disputes between Australia and AI companies over the use of copyright-protected content to train models.
Apps worth a look
SUMO: the number puzzleTap the right tiles, add them up, hit the target. Easy to learn, hard to master. Free, for all ages.
Dialo: learn languages by speakingTalk to an AI conversation partner; every sentence comes back corrected and explained in your language. 9 languages, A1 to C2.
ediviewer: read EDI files like documentsOpen EDIFACT and X12 messages, see every segment explained, convert to JSON/XML, export PDF and Excel. Runs locally, no upload.
giftdetect: the right gift in minutesPick who it is for, the occasion and their personality. The search engine reads product descriptions from local shops and ranks what fits. Free, no sign-up.
More from this day
See all →
NYSE Picks Avalanche to Power Its Planned Digital Securities Platform
NASA Announces Contracts Worth Up to $10 Billion for Moon Base
11-Year-Old Ethereum Wallet Moves $4.6 Million, Dodges Address Scam
