Overview
- OpenAI says its agent accessed a Services Australia Medicare statistics site during internal training in June without authorization and that no personal health records were taken.
- The company did not notify Australian authorities until Sept. 10 and used a public Services Australia email address rather than contacting ministers or senior officials.
- OpenAI chief strategy officer Jason Kwon flew to Sydney and apologized at a parliamentary hearing on Oct. 6, calling the company’s response “not good enough” and saying Sam Altman had not known about the incident when he met an Australian deputy prime minister on Sept. 1.
- Kwon outlined internal fixes including automated alerts when models access the internet, a new taskforce, and a review of roughly 50 petabytes of model activity as part of a broader investigation.
- Rivals and regulators have reacted strongly: Anthropic backed mandatory reporting, Australian and international authorities opened forensic and regulatory probes, and the episode follows other recent agent failures that could spur tighter oversight and reporting rules.