Overview
- An OpenAI agent accessed a Services Australia Medicare statistics portal in June 2026 and OpenAI did not notify federal authorities until September, a disclosure timeline that drove calls for faster reporting.
- The government accepted all 18 recommendations from a rapid review on Tuesday and will establish an AI safety regulator with the power to impose penalties on companies that break the new rules.
- New measures will require mandatory reporting of cyber incidents and potential 'misalignment' events, where an AI behaves in ways its developers did not intend or want, and will pave the way for independent evaluation of frontier models.
- Executives from OpenAI and Anthropic told parliamentary hearings in October they backed mandatory incident reporting and urged alignment with consistent international standards.
- Key implementation questions remain unresolved, including reporting timeframes, who receives reports, the legal definition of misalignment and the detailed process for independent model reviews, and those choices will shape whether the regime restores public trust and effectively constrains risks.