In 2023, Australia published a set of voluntary AI Ethics Principles and initiated consultation on mandatory AI guardrails — a process that generated significant industry and public input and appeared to be building toward binding AI regulation. By 2025, the government had largely abandoned that track, replacing the mandatory guardrails approach with an updated voluntary framework and a sector-by-sector regulatory strategy.
The Australian AI regulatory retreat is one of the most instructive data points in the global AI governance landscape — not because Australia's approach is wrong, but because the reasons for the retreat illuminate the genuine tensions in AI regulation that every jurisdiction is navigating.
What Australia Was Building
The mandatory AI guardrails consultation, led by the Department of Industry, Science and Resources, proposed a set of requirements for organizations developing or deploying "high-risk AI" in Australia. The proposed guardrails were risk-based, broadly aligned with the EU AI Act's domain categories, and included transparency, accountability, human oversight, and impact assessment requirements.
The consultation attracted hundreds of submissions from industry, civil society, academia, and government agencies. By most accounts, the policy development was substantive and the proposals were taken seriously.
What happened between consultation and policy: a combination of factors that reflect real tensions rather than bad faith.
Why It Stalled
Industry opposition framed around competitiveness. The dominant industry narrative in Australia's consultation was competitiveness risk — the argument that mandatory AI regulation would disadvantage Australian companies relative to US competitors operating under lighter-touch federal guidance. This argument has more resonance in smaller economies where the competitive cost of asymmetric regulation is higher and the regulatory capacity to enforce complex rules is more limited.
Regulatory capacity constraints. Australia doesn't have a dedicated AI regulator. The existing regulatory architecture — with the OAIC handling privacy, ASIC handling financial services, TGA handling medical devices, and so on — was the proposed vehicle for sector-specific AI enforcement. The mandatory guardrails approach would have required significant uplift in regulatory capacity across multiple agencies simultaneously, which the government assessed as operationally difficult in the near term.
The "wait and see" on EU AI Act implementation. With the EU AI Act entering its implementation phase, the Australian government made a judgment that observing EU enforcement experience before committing to a similar framework was prudent. This is a defensible policy position — the EU AI Act's compliance burden is still being characterized, and jurisdictions that move second benefit from that experience.
The sector-specific alternative. Rather than a horizontal AI law, Australia committed to working through existing sector regulators to develop AI-specific guidance within their remits — similar to the UK's approach. ASIC on AI in financial services, TGA on AI in medical devices, OAIC on AI and privacy. This approach has precedent (it's working reasonably well in the UK) and is more achievable given existing regulatory capacity.
What Australia Actually Has Now
The National AI Strategy (updated 2025) and the voluntary AI Safety Standard provide the current framework. Key elements:
Voluntary AI Safety Standard. Organizations developing or deploying AI in Australia are encouraged (not required) to apply ten safety standard elements covering governance, transparency, testing, human oversight, and incident response. The standard is closely aligned with the EU AI Act's principles and the NIST AI RMF, which allows organizations with existing compliance programs to apply them without rebuilding.
Sector-specific regulatory activity. ASIC has issued guidance on AI in financial services. TGA has updated its AI as a medical device framework. The OAIC has issued AI and privacy guidance. These are advisory rather than binding in most cases, but they signal enforcement priorities.
International engagement. Australia is active in the Global Partnership on AI, the OECD's AI work, and bilateral AI governance agreements, particularly with the US and the EU. This multilateral engagement is partly compensating for the absence of domestic binding regulation — Australia is shaping international norms even while deferring domestic binding action.
What It Tells Us About Global AI Governance
Voluntary frameworks are not nothing — but they're not enough for high-risk AI. Australia's voluntary approach will produce meaningful governance improvement for organizations that engage with it seriously. It won't produce the baseline of mandatory compliance that protects individuals from organizations that don't engage voluntarily. The sectors where this gap matters most are the same ones the EU AI Act focuses on: healthcare, financial services, employment, law enforcement.
Small-to-medium economies face a structural dilemma in AI regulation. The EU can regulate unilaterally because its market is large enough to impose compliance on global companies. Australia, Canada, Singapore, and similar jurisdictions can influence global norms through voluntary frameworks and multilateral engagement, but they can't unilaterally shape the behavior of global AI companies the way the EU can. The Australia experience illustrates that dilemma clearly.
The "watch the EU and learn" strategy is rational but has costs. Second-mover advantages in regulation are real — learning from EU enforcement experience before committing is reasonable. The cost is that in the gap period, AI-related harms that binding regulation would have prevented are occurring. That tradeoff is being made explicitly or implicitly by most non-EU jurisdictions right now.
For practitioners advising organizations in APAC: the practical compliance picture in Australia is voluntary frameworks plus sector-specific guidance plus EU AI Act obligations for any EU-market operations. The EU obligations are likely to be the most operationally demanding layer and the one that effectively sets the governance floor for multinationals regardless of Australian domestic requirements.
Further reading:
Top comments (0)