The Pulse
Australia’s Governments Adopt National AI Assurance Framework
Commonwealth, state and territory governments have agreed on a national framework for assessing artificial intelligence in public services, setting shared expectations for development, procurement, deployment and oversight.
AI.info Team ·
Commonwealth, state and territory governments agreed on a national framework for assessing artificial intelligence on 21 June 2024, setting shared expectations for how public agencies develop, procure and deploy AI.
Data and Digital Ministers released the framework after meeting in Darwin. The agreement is based on Australia’s AI Ethics Principles and is designed to give governments a common approach without requiring every jurisdiction to use identical procedures.
The framework focuses on assurance: the evidence, oversight and controls agencies use to determine whether an AI system is lawful, safe and suitable for its intended purpose. It covers the full lifecycle of a system, from design and data selection through testing, deployment, monitoring and possible withdrawal.
The ministers’ joint statement sets out a shared approach to the safe and responsible use of AI by Australian governments.
Ministers tie AI adoption to public trust
Senator Katy Gallagher, Australia’s minister for finance and the public service at the time of the announcement, described the framework as a step toward public confidence in government use of AI.
“Today we’ve agreed across all levels of government that the rights, wellbeing, and interests of people should be put first whenever a jurisdiction considers using AI in policy and service delivery.”
Katy Gallagher, minister for finance and the public service
The minister said people should not face unexplained operational differences in government AI systems depending on where they live. The framework therefore aims to create consistent expectations for oversight while preserving room for local legislation, policy and operating requirements.
Ed Husic, then minister for industry and science, said governments would need to use AI to work faster and more efficiently but would also need to maintain public trust in how the technology operates. Jihad Dib, New South Wales minister for customer service and digital government, said the principles-based approach would allow flexibility as AI systems and their risks change.
Five assurance cornerstones guide agencies
The framework identifies five mechanisms, or cornerstones, that governments should consider when building assurance processes: governance, data governance, a risk-based approach, standards and procurement.
Governance requirements include clear responsibility for AI use, updated decision-making structures, risk management and access to technical, legal and social expertise. Agencies are expected to train staff to use AI lawfully, exercise judgement over its outputs, identify problems and explain AI-influenced outcomes.
Data governance receives separate treatment because the quality and representativeness of data affect an AI system’s results. Agencies are expected to manage datasets in ways that meet legal obligations and support reliable, authenticated and appropriate information.
The framework does not prescribe the same level of scrutiny for every use case. It calls for assessments proportionate to the potential harm, with more demanding controls for high-risk systems and a lighter administrative burden for lower-risk applications.
Testing continues after deployment
Assurance is not limited to an approval decision before launch. The framework says agencies should assess risks across design, verification and validation, deployment, operation and monitoring, with new reviews when a system changes or moves between stages.
Recommended practices include pilot studies, performance testing, red-teaming, human validation and continuous monitoring. Agencies should also prepare to disengage an AI system quickly if it causes a data breach, unauthorised access, security compromise or another unresolved problem.
Procurement rules form part of the same process. Government contracts may need to specify accountability, data transparency, access to relevant information, performance testing and evidence that the system continues to operate as expected throughout its lifecycle.
Transparency and appeal are built into the model
The framework says people should be told when an AI system is being used in ways that significantly affect them. Governments should maintain records of an AI system’s purpose, intended uses and limitations, and should retain reliable information about decisions, testing and the data used.
When AI influences an administrative decision, the framework calls for clear explanations and human accountability. Explanations should address the system’s inputs, the factors that influenced its reliability, test results and the role of human oversight.
People affected by an AI system should also have a timely way to challenge its use or outcome. Governments are expected to communicate rights and protections clearly, provide channels for objections and complaints, and ensure access to human review, dispute resolution or investigation where appropriate.
A national baseline, not a single operating manual
The framework is intended to sit above individual federal, state and territory policies. The Department of Finance says jurisdictions can develop more specific rules that reflect their own laws, administrative structures and operational conditions.
Its eight underlying AI Ethics Principles cover human, societal and environmental wellbeing; human-centred values; fairness; privacy protection and security; reliability and safety; transparency and explainability; contestability; and accountability.
That arrangement gives agencies a shared baseline while leaving them responsible for deciding how much assurance a particular system requires. The framework also expects governments to exchange lessons from their own systems and improve their assurance practices as new risks and uses emerge.