National framework for the assurance of artificial intelligence in government

What it is

The national framework for the assurance of artificial intelligence in government is a joint approach to safe and responsible AI by the Australian, state and territory governments. It was agreed and released by the Data and Digital Ministers Meeting on 21 June 2024.

The framework does not focus on technical detail. It sets foundations across government, and each jurisdiction develops specific policies and guidance for its own legislative, policy and operational context. It is based on Australia’s AI Ethics Principles and sets out five cornerstones of assurance and practices for applying each principle.

The framework describes assurance as what lets governments understand the expected benefits of AI, identify risks and apply mitigations, ensure lawful use, understand whether AI is operating as expected and demonstrate, through evidence, that the use of AI is safe and responsible.

Who it applies to

Australian, state and territory governments and their agencies, through each jurisdiction’s own implementation. Private organisations are not bound. Vendors to government feel it through procurement.

The five cornerstones

1. Governance. Organisational structures, policies, processes, roles, responsibilities and risk management for AI, adapted from existing decision-making and accountability structures and proportionate to risk. Leaders should commit to safe and responsible use and build a positive AI risk culture, giving staff the training and resources to use AI ethically and lawfully, exercise judgement over outputs and explain AI-influenced outcomes.
2. Data governance. Policies, processes and roles that keep datasets authenticated, reliable, accurate and representative, with robust data governance that complies with relevant legislation.
3. A risk-based approach. Assess and manage AI case by case, with more scrutiny in high-risk settings and minimal burden in lower-risk ones. Manage risks through the lifecycle, with reviews at transitions between phases, monitoring and feedback loops, plans for obsolete systems and oversight bodies such as AI risk committees for high-risk settings. The framework cites self-assessment models such as the NSW AI Assessment Framework.
4. Standards. Align with relevant AI standards where practical, naming AS ISO/IEC 42001, 23894 and 38507.
5. Procurement. Address ethics principles, accountabilities, data transparency, access to information and proof of performance testing in procurement documents and contracts. Keep contracts adaptable, build internal skills to avoid vendor lock-in and check whether standard clauses cover AI’s new and amplified risks.

Applying the principles

Beyond the cornerstones, the framework sets out practices for each of the eight AI Ethics Principles, showing what each one means for government assurance in practice. It is designed so that jurisdictions add learnings to future iterations as their assurance practices mature.

What it means in practice

For an agency, the cornerstones read as a set of questions an assurance approach should be able to answer.

• Who is accountable for each AI use, and do existing governance bodies have sight of it?
• Is the data behind each system reliable, accurate and representative, and is it governed under the relevant law?
• Has each use been assessed in proportion to its risk, and is it reviewed when it moves between lifecycle phases or changes significantly?
• Which standards does the agency align with?
• Do contracts give the agency the information, testing evidence and vendor support it needs to monitor a system and respond to an incident?

For a vendor selling AI into government, the procurement cornerstone is the one that bites. Agencies are told to ask for clear accountabilities, data transparency, access to information and proof of performance testing across the lifecycle. They are also told to check whether a vendor can support review, monitoring and evaluation of outputs after an incident. Vendors that can supply that evidence make an agency’s assurance work possible.

Key dates

The framework was agreed and released on 21 June 2024. Victoria’s generative AI guideline records that the Victorian Government agreed to it at that meeting. The Commonwealth’s DTA policy was revised to version 2.0 in December 2025, and the Queensland and NSW frameworks were reissued in 2026. Queensland’s AI governance policy points agencies back to the national framework.

How it relates to other frameworks

The framework sits between the AI Ethics Principles and each government’s own instruments. Those instruments are the DTA policy for the Commonwealth, the NSW AI Assessment Framework, the Queensland AI governance policy and FAIRA, the WA AI Policy and Assurance Framework and the Victorian generative AI guideline. Its standards cornerstone points to ISO/IEC 42001, ISO/IEC 23894 and ISO/IEC 38507. For businesses, the closest equivalent is the voluntary Guidance for AI Adoption.