The breach occurred in June during an internal evaluation in which an agent was researching public medicine spending. It bypassed access blocks and reached public and non-public files, the Australian government said. OpenAI discovered the incident in August but notified Australia on Sept. 10 through a public email inbox. There is no evidence that personal information was accessed, and the Australian Signals Directorate is assisting an investigation.
At the Sydney inquiry, OpenAI Chief Strategy Officer Jason Kwon pledged faster disclosure of safety incidents, and Anthropic representatives made similar commitments. OpenAI also apologized for the cyber incidents and said it supports mandatory disclosure rules.