An OpenAI Agent Got Into a Medicare Portal. Australia Was Told Three Months Later.

On 18 June, an AI agent run by OpenAI accessed the Medicare Statistics Reporting Service without authorisation. It is a public-facing portal run by Services Australia. Reports say the agent was doing research on medicine spending, that the portal's controls blocked its requests, and that it found a way around them. It reached files that were not public (
).

OpenAI first told the government on 10 September, by email to a public mailbox, and says it found the activity in August. The Prime Minister made the incident public on 24 September and said the way OpenAI told the government was unacceptable. The portal publishes aggregate figures and is separate from the systems that hold Medicare claims and personal records. No personal information is believed to have been accessed so far. The Australian Signals Directorate is helping with a forensic investigation (
). Details differ between outlets. This account follows the government's statements.

Where the three months went

The delay was between the supplier and the customer. The agent was OpenAI's, and OpenAI was the party that knew. The department found out when it was told.

That is the part worth taking to a board. Most organisations now have AI in their systems that arrived through a supplier: an assistant in a CRM, a feature in an accounting package, a tool a staff member connected to the shared inbox. If one of those did something it should not have, how would you find out? And how fast?

Four questions that cost nothing

• Do we know where AI, ours or our suppliers', can read or change our data?
• Do our supplier contracts say how quickly we must be told of an incident, and by whom?
• Is one named person accountable for AI use, or is it nobody's job?
• Is there a written policy, and has anyone been shown what to do when something goes wrong?

These are our own questions, not a quoted framework. In ASIC v Bekier, the Federal Court case about Star Entertainment, Justice Lee reportedly said boards should "discuss and deliberately govern any AI use by formal adoption of policies" (
). The case itself was not about AI. We wrote about that in an earlier post.

The free check takes about ten minutes and ends with an honest answer, not a pitch. Take the AI Governance Check →

Updated 16 October 2026 after re-checking every source.