Introduction
The AI Reference Architecture (AIRA) provides an architectural foundation for designing, implementing and governing artificial intelligence (AI) across the Australian Government. It supports the implementation of the AI Plan for the Australian Public Service 2025 and the Policy for the responsible use of AI in government.
As agencies increasingly adopt AI, AIRA provides a common set of structures and taxonomies to manage complexities and costs, support interoperability, promote reuse and enable more consistent decision making, governance and investment.
Grounded in established ICT architecture practices, AIRA promotes modular and scalable AI solutions, while embedding security, privacy and ethical considerations. It helps agencies manage the complexity of AI adoption across technologies, operating models and governance arrangements, supporting resilient and coherent AI ecosystems.
Developed in consultation with government agencies, AIRA is designed to support a range of AI adoption pathways and maturity levels, from vendor managed services to bespoke applications and internally developed models. It applies across machine learning, generative AI, agentic AI and emerging frontier AI capabilities, providing a flexible framework that agencies can adapt to their business objectives, operating environment and risk profile.
Using AIRA in your agency
AIRA complements existing whole-of-government policies, standards and guidance. It provides a common basis for the development of architectures for AI-enabled systems, helping agencies address policy obligations and standard requirements in a consistent way. See Appendix B: Relationship with other frameworks for details on how AIRA relates to government policies, standards, industry architecture frameworks and operational frameworks.
In practice, agencies can use AIRA to:
- Architect AI systems
- assess current-state architecture and shape target-state architecture
- identify dependencies, integration points and specialist capabilities required for delivery
- scope solutions and produce more consistent first-pass effort and complexity estimates
- Inform procurement decisions
- inform buy-build-blend and reuse decisions
- support consistent, like-for-like evaluation of vendor offerings
- Govern AI effectively
- support agencies in assessing readiness to introduce, scale and retire AI systems safely, with appropriate controls, oversight and auditability
- provide input to an AI Use Case Impact Assessment by helping agencies understand how architectural components, controls and dependencies influence where risk resides across an AI system
- mitigate impacts identified from AI Use Case Impact Assessment through architectural controls and assurance activities. This includes additional evaluation criteria, design-time controls embedded in system architecture, as well as run-time controls such as monitoring and human oversight mechanisms.
- trace and implement requirements from the AI technical standard, including security, privacy, human oversight and safety controls
- assign clear responsibilities across the architecture
- structure and locate assurance evidence across the architecture
- Optimise investment
- facilitate reuse, which may help reduce delivery effort and accelerate time to value
- inform AI cost modelling and return-on-investment analysis.
AIRA is intended to be used after the problem has been defined and validated. AIRA does not replace problem-framing guidance, business case development, procurement documentation, return-on-investment calculation or detailed implementation guidance.
Demonstrating AIRA in practice
This is accompanied by the AI Reference Architecture Companion Guide - Scenarios which provides examples that demonstrate how AIRA can be used across common government AI adoption pathways. Each scenario provides a high-level overview and is accompanied by a detailed architectural analysis, capability decomposition, trustworthiness considerations and human oversight patterns.
The scenarios provide a starting point for practitioners to use AIRA in structuring decisions, identifying dependencies, exploring trade-offs, understanding where responsibilities reside and determining how system behaviour can be meaningfully shaped, constrained, monitored and controlled across an AI-enabled system. Throughout, AIRA reinforces the importance of understanding the problem and establishing an evaluation approach before architecting solutions and selecting technologies. These disciplines help agencies identify the simplest architecture capable of meeting business needs and introduce complexity only where supported by evidence.
Audience
The primary audience for AIRA includes roles responsible for design, build, deployment, operation and oversight of AI enabled systems.
This includes:
- architects across business, information, application, technology and security domains
- engineering, platform, data and AI application development teams
- delivery and operational functions
- governance, assurance and compliance practitioners.
Leadership and accountable officials use AIRA to support strategy, investment, governance and risk decisions, while procurement and information management functions use it to promote interoperable sourcing, consistent recordkeeping, provenance and stewardship of AI systems and artefacts.
Other audiences include groups that may benefit from AIRA but are not expected to use it directly. This includes the Australian public and businesses, who benefit from more consistent, accessible and reliable digital services across government, as well as industry, suppliers, regulators, oversight bodies and the research community. For these audiences, AIRA provides transparency into the Australian Government’s approach to AI and helps align external capabilities, services and investments with whole of government architectural expectations.
Getting started
To understand AIRA’s structure and core concepts, start with the Architecture overview, then read the Architecture description.
To apply AIRA to a specific AI adoption challenge, refer to AI Reference Architecture Companion Guide - Scenarios, and choose the example that is closest to your situation.
To explore how different roles can use AIRA, see Appendix A: Key roles in AIRA.
To understand how AIRA relates to other government and industry frameworks see Appendix B: Relationship with other frameworks.