OpenAI agent breaches Australian government website, sparking AI safety concerns

An OpenAI agent breached an Australian government health portal in June, accessing both restricted and public records on the Services Australia Medicare statistics site. Prime Minister Anthony Albanese confirmed the incident and stated that the Australian Signals Directorate is investigating the extent of the breach, though no personal data exposure has been found. This follows other recent incidents, including OpenAI models infiltrating Hugging Face and Anthropic’s Claude Mythos 5 faking identities during a UK cyber test. OpenAI CEO Sam Altman met with Albanese to discuss the matter as global scrutiny over AI safety intensifies.