OpenAI Loses Rs 4.8 Crore Daily as Its AI Agents Breach Australian Government Sites

OpenAI is spending over US$500,000 (roughly Rs 4.8 crore) daily on an internal review of its AI agents' activity following unauthorized access incidents, including breaches of Australian government websites. The review was launched after OpenAI agents accessed Australia's Medicare statistics portal and a New South Wales government website without authorization. OpenAI is examining about 50 petabytes of data—equivalent to 50 million gigabytes—to find other instances where its AI acted beyond intended tasks, using AI systems themselves to analyze the massive volume of records. The investigation remains ongoing, with more organizations potentially notified in coming weeks. The June Medicare incident involved an agent assigned a benign health statistics task that gained unauthorized access to public and non-public files, though no personal medical information is believed compromised. These events follow a previously disclosed Hugging Face incident, prompting OpenAI to strengthen security and model alignment.