Australia's AI Safety Institute

What Australia’s AI Safety Institute does

Australia’s AI Safety Institute has 3 goals:

• Analyse and test new AI models and applications.
• Support regulators and agencies in responding to emerging AI risks and harms.
• Shape safe AI development, deployment and international governance in Australia's interests. 

Its work contributes to keeping Australians safe from AI-enabled risks and harms. 

It supports work across government to:

• ensure that Australian laws and agencies continue to protect Australians from AI-enabled harms
• promote responsible use of AI
• engage internationally to support AI development that is in Australia’s interests.

Supporting government to understand emerging AI capabilities

Australia’s AI Safety Institute supports government and regulators’ understanding of emerging AI capabilities, risks and mitigation approaches. It has in-house technical specialists and works with technical partners including the Australian Signals Directorate and CSIRO. It also works with domestic and international experts.

Research into frontier model capabilities

Australia’s AI Safety Institute monitors and analyses emerging AI capabilities, with a focus on advanced and frontier AI models. This includes systems such as AI agents that can act autonomously to pursue goals, and highly capable generative models that produce text, images, video and other content.

This work helps government anticipate the new systems and platforms – and emerging risks – that new AI capabilities will generate. It also provides an evidence base to inform future policy and governance decisions.

Monitoring real‑world harms and emerging risks

Australia’s AI Safety Institute monitors how AI is used in our society and our economy. It will gain important insights into the actual effects of AI technologies on families, communities, businesses and society more broadly. 

Those insights will help regulators and policy makers understand the practical effects of AI as they emerge.

Drawing evidence from domestic and international expertise

Australia’s AI Safety Institute engages with domestic and international experts to stay informed, credible and connected as global approaches to AI safety continue to develop. This includes:

• Actively participating in the International Network for Advanced AI Measurement, Evaluation and Science (NAAIMES), formerly called the International Network of AI Safety Institutes, to work on strengthening the science that underpins AI evaluation.
• Sponsoring Australia’s AI Safety Forum, an interdisciplinary forum grounded in the science of AI safety. It brings together people from across research, government, industry and civil society to share information on the latest in AI safety science.
• Collaborating on AI safety and security under memorandums of understanding with frontier AI labs.

Joint testing exercises 

Australia contributes to joint testing of frontier AI systems through the NAAIMES. These exercises aim to improve our ability to accurately measure AI capabilities and risks. This will help to better identify, understand and manage the risks of AI systems before they cause harm to Australians. 

Australia has contributed to several joint testing exercises, drawing on expertise from Australian researchers, to:

• improve how countries test foundation models 
• check how AI works across different languages 
• assess risks from AI agents such as data leaks, fraud and cyber threats.

Supporting regulator and agency responses to emerging AI‑related risks and harms

Australia’s AI Safety Institute will provide technical information and insights to support regulators and agencies so they can keep pace with AI developments. 

Australia’s AI regulatory approach builds on strong labour market institutions and effective, largely technology‑neutral laws across areas such as:

• privacy
• consumer protection 
• online safety 
• workers’ rights
• anti‑discrimination.

Australia’s AI Safety Institute supports a coordinated and coherent approach to addressing AI-harms and risks across portfolios. 

Shaping safe AI development, deployment and governance in Australia’s interests

Australia’s AI Safety Institute works with partners and experts around the world to keep pace with the latest technical research and best-practice approaches to AI governance. 

Domestic engagement and partnerships

Australia’s AI Safety Institute engages with agencies and regulators on AI capabilities, risks and harms to inform AI governance and best practice. 

Australia’s AI Safety Institute engages with technical experts to support safe AI deployment aligned with Australian laws, regulation, values and interests. These include:

• domestic technical experts
• Australian Signals Directorate
• CSIRO 
• National Intelligence Community 
• tech companies and international partners.

Australia’s AI Safety Institute partners closely with the National AI Centre (NAIC), which leads engagement with industry and the public on practical guidance for safe and responsible AI adoption. The NAIC translates the AI Safety Institute’s insights into practical guidance for businesses, organisations and the public.

International engagement and influence

Australia is a founding member of NAAIMES. Australia signed the Seoul Declaration in May 2024, confirming a commitment among countries to advance the science of AI safety, building on the Bletchley Declaration.

Australian AI experts have contributed to the independent International AI safety report. Over 100 independent experts contributed to the report, including an Expert Advisory Panel nominated by more than 30 countries.

Through engagement with NAAIMES and the International AI Safety Report, Australia’s AI Safety Institute helps ensure Australia remains informed, credible and connected as global approaches to AI safety continue to develop.

Advancing AI safety science

AI safety science is the study of how to reliably measure, understand and shape the behaviour of AI systems, and the impacts they produce in the wider world.

Australia’s AI Safety Institute contributes to that agenda by conducting, supporting and participating in research that improves understanding of AI capabilities, risks, harms and mitigations. This includes research focused on:

• how advanced AI systems behave in complex or multi‑agent environments
• how risks evolve as AI systems interact with people, organisations and each other
• how technical evaluation and measurement methods can better detect emerging risks.

This research strengthens the evidence base available to government and informs future policy and governance decisions.