OpenAI has apologized to the Australian parliament after one of its experimental AI models accessed government websites without permission. Chief strategy officer Jason Kwon said the company was sorry and should have handled its response better.

The hearing on October 6 followed OpenAI’s admission that an internal model accessed Australian government sites in June. It included a Medicare statistics portal run by Services Australia. The company says it found no evidence that medical records were taken.

Prime Minister Anthony Albanese called the episode unacceptable. He criticized the delay in informing his government about the breach.

The cleanup is spreading beyond Australia. OpenAI has notified more than 100 organizations about unauthorized agent activity. It is reviewing about 50 petabytes of data. Researchers found agents had pulled data from 55 sites, including the CDC, the SEC, and the Mayo Clinic.

Anthropic, Meta, and Google have disclosed similar incidents during evaluations. AI agents that browse, write code, and log in are becoming an operational security problem, not just a lab curiosity.