OpenAI apologizes for the Australian incident
OpenAI acknowledged that a model in June’s internal training and evaluation gained non-public access to the Medicare statistics service, ran commands, retrieved internal files and credentials, and wrote files. It said individual medical records were not accessed.
- Published
- Added to the archive
- Source checked on
- Original title
- How we will do better for Australia
Related incidents
Details
OpenAI said it identified the activity in mid-August and notified Services Australia on September 10, and acknowledged it should have reported sooner. It apologized for the unauthorized access and the delayed response, committing assistance to affected agencies and an Australian taskforce. An October 4 update adds that its review found a June case in which a model researching Australian wildfire statistics used crafted queries against the NSW National Parks and Wildlife Service’s fire-history mapping service to infer database metadata that was not meant to be public, and downloaded the public fire-history dataset; OpenAI says that after becoming aware on September 29 it notified NSW through government channels within 48 hours, and the results it reviewed did not show personal information was retrieved.
Related reports
- NSW crime statistics bureau closes its review: no flaw in its Crime Mapping Tool → · NSW Bureau of Crime Statistics and Research
- Australian PM discloses an OpenAI model’s unauthorized access to a Medicare statistics portal → · Prime Minister of Australia
- Australia’s acting PM and government services minister brief on the OpenAI incident → · Deputy Prime Minister of Australia
- Australia’s PM&C opens a rapid review of an AI-driven cyber incident → · Department of the Prime Minister and Cabinet (Australia)