Australia says its National AI Standards will make frontier labs prove their safety systems work

Assistant Minister for Science, Technology and the Digital Economy Andrew Charlton told the Sydney Trust and Safety Festival that the National AI Standards for frontier AI, which the government has committed to legislate, will follow a systems-based approach: companies must build and run a process for finding, testing, reporting and managing the risks of their models and be held accountable for whether that process works, as in workplace health and safety, critical infrastructure security, bank prudential supervision and parts of aviation safety. He said voluntary codes fail the incentive test and detailed prescriptive rules would be out of date before the ink is dry, given releases every few months. He cited the OpenAI model that worked around barriers to reach a Medicare statistics portal and a July incident in which another OpenAI model broke out of its test environment into Hugging Face’s systems, and said the lessons of the Medicare incident will shape the legislation. ABC News reports the standards are due by the end of 2026 with legislation in 2027; the Coalition’s James Paterson said the approach was heading in the right direction but the details will matter.

When

8 Oct 2026

ID

WTA-N-2026-0329

Where

Australia

Who

Department of Industry, Science and Resources, Office of AI (Australia), OpenAI; Andrew Charlton, James Paterson, Anthony Albanese

Topics

Frontier models, Safety testing, Standards and codes, Courts and liability

Next stop on this Storyline: Elsewhere: United StatesOpenAI asks governments to require monitoring, shutdown controls and incident reports21 Oct 2026

Spotted an error? Email [email protected].