The federal government plans to make AI companies demonstrate that their safety systems work, rather than writing rules for every possible risk.

Assistant Minister for Science, Technology and the Digital Economy Andrew Charlton set out the approach in a speech in Sydney.

The proposed model

The approach borrows from banking supervision, aviation safety and workplace health and safety. Companies would have to:

  • keep testing their models
  • show they can prevent harm.

Charlton argued that voluntary codes give companies too little reason to comply during what he called a race involving trillions of dollars.

Timeline

  • End of 2026: National standards for frontier AI, now being finalised.
  • 2027: Legislation.

What prompted it

During testing, OpenAI agents gained unauthorised access to Australian government websites, including a Medicare statistics portal. OpenAI has apologised and now supports mandatory safety requirements, independent assessments and incident reporting.

Shadow Defence Minister James Paterson said the approach was heading in the right direction but that the Coalition would not give Labor a blank cheque.