Multinational enterprises (MNEs) increasingly implement agentic artificial intelligence (AI) to perform value-creating functions previously carried out by human personnel. The OECD Transfer Pricing Guidelines (hereinafter: OECD TPG or Guidelines), however, follow a human-centric and enterprise-centric approach: they allocate risks and decision-based functions to entities through their human personnel and recognise only enterprises as bearers of those risks and functions. As agentic AI at high autonomy levels progressively replaces human personnel in the performance of specific decision-making – the mechanism through which the Guidelines allocate risks and decision-based functions – an “allocation gap" emerges: contributions to value creation by agentic AI within MNEs can be allocated – and consequently taxed – neither at the level of an enterprise nor the agentic AI itself. This allocation gap is structurally embedded in the Guidelines, thus affecting all areas of transfer pricing where agentic AI is implemented within an MNE and all cross-border transactions between its associated enterprises. Importantly, this conclusion holds regardless of whether the agentic AI is operated under human governance, such as a “human-in-the-loop".
