Australia’s emerging AI governance reality for law, policy and practice
About this presentation
These slides were originally developed to accompany a presentation delivered at Maddocks in April 2026. As a result, the slides are intentionally visual and contain only a portion of the argument. Much of the explanation, context, and analysis was delivered verbally.
The presentation examines Australia’s emerging AI governance landscape following the release of the National AI Plan. Rather than reviewing the Plan itself, it asks what happens next. If Australia has chosen to govern AI largely through existing institutions, courts, regulators, procurement processes, professional obligations, organisational governance, and technical evaluation, what new responsibilities does that create for those institutions?
The embedded slides below contain the visual material used during the presentation. An abridged written summary follows for readers who prefer text.
Presented at Maddocks Law firm in April 2026
Abridged Summary of my speaking notes.
Australia’s AI governance challenge is increasingly institutional
The National AI Plan establishes direction for AI adoption, capability, investment, and safety. However, it does not function as a dedicated AI law. Instead, governance responsibility is distributed across existing institutions including courts, regulators, public agencies, professional bodies, procurement processes, and organisational governance structures.
The governance handoff is already underway
Developments such as the Federal Court’s Generative AI Practice Note demonstrate that AI governance is already becoming part of professional practice. Organisations and practitioners are increasingly expected to understand what AI systems were used, how they were used, and who remains accountable for outcomes.
Existing law can identify harms but may struggle to identify trajectories
Many legal frameworks are designed to address outcomes such as privacy breaches, discrimination, misleading conduct, or procedural failures. Agentic AI systems introduce an additional challenge: understanding the chain of interactions that produced those outcomes. The path that generated a decision may become as important as the decision itself.
Agentic systems change the governance object
Traditional AI governance often focuses on the model. In practice, organisations deploy configurations consisting of models, prompts, retrieval systems, tools, memory, interfaces, workflows, and human oversight arrangements. Two organisations can deploy the same foundation model while creating entirely different governance risks.
Human oversight must be meaningful
A human in the loop is not automatically a governance control. Effective oversight requires visibility into system behaviour, authority to intervene, sufficient time for review, and enough understanding to exercise judgment. Without these conditions, human review can become a procedural formality rather than a genuine safeguard.
Evaluation is part of governance
AI governance is not only a matter of law, policy, and compliance. It also depends on what organisations choose to measure. Testing, monitoring, auditing, incident reporting, documentation, and benchmarking help determine which risks become visible and which remain hidden. Evaluation therefore functions as a governance mechanism rather than a purely technical activity.
Australia’s missing layer is evaluative infrastructure
If Australia intends to govern AI through distributed institutional responsibility rather than a comprehensive AI Act, those institutions require practical mechanisms for observing, testing, tracing, and contesting AI behaviour. The critical challenge is not simply establishing rules, but building the evaluative infrastructure needed to make governance operational in practice.
Speaking, Workshops and Advisory Support
I regularly speak to legal, government, industry, and academic audiences on AI governance, evaluation, generative AI, agentic systems, and Responsible AI implementation.
If your organisation is exploring AI governance, developing Responsible AI capability, evaluating AI systems, or looking to better understand the implications of emerging AI technologies, please get in touch.