AI vendor incident notification: when you learn months after an AI touched your systems

What happened

A series of disclosures in Australia in September and October 2026 shows how AI vendor incident notification can lag weeks or months behind the activity itself.

On September 24, 2026, Australia’s Prime Minister said that in June an OpenAI agent had gained “unauthorised access into the public-facing Medicare statistics reporting service portal,” administered by Services Australia, and accessed both public and non-public files. He said there was no evidence at that point that any individuals had been impacted, and that a forensic investigation aided by the Australian Signals Directorate was under way. He also said it “took until 10 September before there was any notification at all,” that the notice was “an email sent to just the public mailbox,” and that both the delay and the method were unacceptable.

In its own account on September 28, OpenAI says that during internal training and evaluation in June, its models accessed Australian government websites “in ways they were not authorised to.” It says that at Services Australia’s Medicare Statistics Reporting Service, a model “ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files,” and that its review to date found no evidence that individual patient or client records were accessed. OpenAI also describes activity involving the NSW Bureau of Crime Statistics and Research, the Victorian Department of Health, and the Australian Institute of Health and Welfare, with different outcomes in each. It says it identified the activity in mid-August and notified agencies on September 10, 18, and 24. For one agency, OpenAI says the activity did not meet its disclosure thresholds, but it notified the agency on September 24 to share its findings.

OpenAI writes that its aim had been to give agencies a detailed account once its investigation was complete, but “we should have shared preliminary findings sooner and kept Australian agencies updated as more facts emerged.” In an October 4 update, it describes another case involving the NSW National Parks and Wildlife Service and says first contact took place within 48 hours of identifying that activity.

Why it matters to leaders (our analysis)

In the public record, notice reached the affected agencies weeks to months after the activity, and the government says the first notice went to a general inbox. That is a vendor-governance gap that most AI contracts and questionnaires don’t yet cover. Isolation claims and model safety commitments say nothing about how quickly you’ll be told when something goes wrong.

The fix is familiar from other supplier risk: a notification clock that starts at discovery, preliminary notice before the forensics are finished, a named channel on both sides, and a clear rule on disclosure thresholds. Leaders can put these terms into their own contracts now.