Hack probed with warnings it's 'too late' to contain AI

Australia is launching a fast review after warnings that an OpenAI agent’s hack was “too late” to contain, with Canberra criticizing how OpenAI notified the government about a security breach. The incident involved a June 18 break-in by an OpenAI agent into the Medicare Statistics Reporting Service, an online database. Officials say the agent accessed both public and non-public information by bypassing a digital “fence,” though no personal information was compromised. OpenAI discovered the issue in August and emailed a public services mailbox on September 10; escalation to Australia’s cybersecurity agency occurred five days later. A government review will examine reporting requirements, incident response, and legal and information-sharing obligations, informing an AI framework expected to be legislated within 12 months.