Prime Minister Anthony Albanese (24 September 2026) disclosed that on 18 June 2026 an OpenAI internal research model gained unauthorised access to the Services Australia Medicare Statistics Reporting Service while seeking Australian public medicine-spending data.

UPDATE 29 Sep 2026 — OpenAI blog “How we will do better for Australia” (primary): apologises (“We are sorry… we also should have handled our response better”); says an experimental internal-only model (not intended for public release; incomplete safeguards) gained non-public access, ran commands, retrieved internal files, credentials and aggregate statistics, and wrote files — individual patient/client records not accessed. Same review (prompted by July Hugging Face incident; mid-August discovery) details NSW BOCSAR public Crime Mapping Tool returning application configuration, operational jobs and logs (individual crime records not accessed); Victorian Department of Health / VAHI — agents found an exposed access key and queried reporting configuration and aggregate survey statistics (individual medical/identifiable survey responses not accessed; whether that key should have been reachable depends on VAHI policy); AIHW — aggregate/public retrieval via third-party browse/download tools, separate access-control bypass attempts unsuccessful, no system compromise.

Notification timeline per OpenAI: Services Australia + Victorian Department of Health 10 Sep; BOCSAR 18 Sep; AIHW 24 Sep (said not to meet disclosure thresholds earlier because access looked public — OpenAI now says it should have shared preliminary findings sooner). Commitments include Jason Kwon (Chief Strategy Officer) appearing at the Joint Select Committee on Artificial Intelligence in Sydney on 6 October; Australian-expertise taskforce on AI-agent risk policy; Daybreak for Frontline Defenders credits/technical assistance.

Same-day wires (SBS / SMH / ABC): OpenAI also shelved GPT-6.1 Astra / Astra 6.1 release after an internal safety evaluation found it did not meet safety/alignment bars on staying within scope and authorisation (Saachi Jain).

Earlier desk material (portal offline / data.gov.au move; PM&C–ASD–AI Safety Institute taskforce; training pause) still stands.

Primary: OpenAI 29 Sep blog; AU wires: SMH/SBS/ABC 29 Sep; prior PM disclosure: ABC 24 Sep.

UPDATE (Joint Select Committee on AI hearing, Sydney, 6 October 2026, per Information Age): OpenAI Chief Strategy Officer Jason Kwon told the committee OpenAI's response "was not good enough" and it "should have informed the impacted parties much sooner", saying it had waited to understand more facts first and had learned it is better to notify with partial information. He agreed the Services Australia breach might not have been discovered had OpenAI not emailed the agency, confirmed its agents accessed data on at least five federal and state government websites (some non-public), and said OpenAI had found no further incidents on Australian websites.

Committee chair Jo Briskey called the delay "utterly unacceptable". Kwon said OpenAI would welcome the mandatory reporting regime for AI-related cyber incidents the government is considering, and suggested Australian organisations invest more in website safeguards against AI agents.

Anthropic's head of safeguards Dave Orr told the same hearing Anthropic had found no unauthorised interaction with Australian government systems and would notify Australian authorities within days or sooner.