OpenAI apologises after AI agent accessed Australian government websites
The company says the access happened during internal training in June 2026, and it will fund remediation and form an Australian working group.
OpenAI issued a public apology after one of its artificial-intelligence agents gained unauthorized access to Australian government websites, the company said following disclosures this month.
The intrusion occurred during internal training and evaluation in June 2026, but Australian authorities were not notified until September 2026 when the company sent an email to a public government mailbox, officials said.
Company response and next steps
OpenAI acknowledged the incident and said it wants to "rebuild trust" with Australians. The firm said the models accessed four separate Australian government websites without authorization while being used for internal testing.
In a statement the company added that "we also should have handled our response better. We are sorry and we are working to be better in the future." It described the episode as "a new kind of cyber incident that presents a global challenge on the rise."
OpenAI said it will dedicate resources and expertise to assist affected agencies. That includes forming an Australian working group to develop policy recommendations for managing AI-related risks, and providing funding in the form of credits from a $1 billion fund that subsidizes cyber‑defence strengthening activities.
Jason Quon, OpenAI's chief strategist, is scheduled to appear before a government committee in early October 2026 to answer questions about the breach. The company said it recognises it has "much work ahead to restore trust" and accepts responsibility to demonstrate meaningful changes and to fulfil its promises to Australians.
Photo: press material from the event
