Much is being written about AI in terms of how we practise tax and the impacts on our roles, but less focus seems to be placed on what it will mean for the rules that govern tax practice. 

The Australian tax system, and the international system within which it sits, are both inherently human-centric frameworks which have largely focused on physical presence. However, in a world in which more decisions are made by AI and greater value is generated through that AI, these frameworks may be difficult to apply and are likely to give rise to anomalous results. 

As AI is inherently mobile, it could result in significant shifts in revenue bases or, more likely, different countries advocating for new attribution models. 

The challenge: AI does not fit traditional tax concepts

International tax law has long relied on the physical presence of people making decisions and performing functions. Yet AI systems can now operate, learn, and generate value with minimal or no human intervention, creating a conceptual mismatch with established rules on permanent establishment (PE), transfer pricing, and tax residency. 

Although Australia is investing in data centres to assist in the growth of AI, the US continues to dominate this space with 45–54% of data centre infrastructure based in the US. Australia has 268 data centres compared to the 3,994 based in the US. 

If an AI system located on a server in the US autonomously manages customer relationships, optimises pricing, or develops new algorithms for a services firm based in Australia, where should the resulting profits be taxed? Whilst some of this debate is not new, it is the value added by, and role played by, servers in these data centres that has changed. Current tax frameworks struggle to answer the question of where profit should be taxed because they were designed around human decision-makers, not autonomous systems. 

Australia's current approach to IP migration

In Practical Compliance Guideline (PCG) 2024/1, which deals with the migration of intangibles, the ATO sets out a risk-based compliance framework for cross-border transfers of intangible assets and related DEMPE (Development, Enhancement, Maintenance, Protection, and Exploitation) functions between related parties. Critically, the guidance covers not only outright asset transfers but also arrangements that have a "similar effect", including potential mischaracterisation or non-recognition of Australian DEMPE functions connected to offshore-held intangibles. 

AI systems — including algorithms, models, training data, and the systems themselves — clearly constitute valuable intangible assets under transfer pricing principles. In PCG 2024/1 the ATO sets out the contemporaneous documentation and evidence it may require to clarify DEMPE activities. 

However, these requirements assume human decision-making and physical performance. What happens when decisions and functions are performed solely by machines? The rate of change at most businesses is exponential, with functions being redesigned constantly and an increasing push to automate and shift value creation to AI. There will be inevitable migration of intangibles which will often be difficult to identify — and even more difficult to document to the satisfaction of a revenue authority. 

The residency question: where is AI "managed and controlled"?

Under Australian law, a foreign-incorporated company can be treated as an Australian tax resident if it carries on business in Australia and has its central management and control (CM&C) in Australia. CM&C focuses on where high-level policy and strategic decisions are made as a matter of substance — not merely where they are formalised or where directors happen to reside. 

Australian boards and executive management will increasingly look to AI to improve their decision-making, and as a result it is expected that  some high-level policy and strategic decisions will move to AI. If these functions constitute the actual exercise of CM&C of business operations — rather than merely routine operational matters – a foreign company could be argued to have CM&C in Australia, triggering Australian tax residency. 

Practical implications and risk mitigation

For international tax managers, several key actions are recommended: 

1. Map your AI-related DEMPE functions: Document where genuine development, enhancement, maintenance, protection, and exploitation of AI systems occurs. Ensure that intercompany agreements, transfer pricing documentation, and actual operations are aligned, with clear evidence of who controls economically significant risks and performs important functions.
2. Review decision-making protocols: Assess whether key strategic decisions are genuinely being made offshore or whether the substance of decision-making (including through AI systems) is occurring in Australia.
3. Assess transfer pricing positions: Given the ATO's focus on intangibles migration and DEMPE functions, ensure your transfer pricing policies properly compensate Australian entities for AI-related functions performed locally, with contemporaneous documentation supporting those positions.
4. Monitor evolving guidance: International tax rules are struggling to keep pace with AI development. Tax managers should monitor global guidance and developments and expect increased scrutiny and potential rule changes. 

AI represents a fundamental challenge to international tax frameworks built around human decision-making and physical presence. For multinationals with Australian operations, the ATO's existing rules on corporate residency and intangibles migration provide a glimpse of how tax authorities will approach AI-related profit allocation – even in the absence of AI-specific legislation. Contact MinterEllison's tax team to understand how these developments may affect your business.