Research Summary: Welfare-Opaque Income: Taxation under AI-Agent Delegation

Research identifies 'welfare-opaque income' as a novel challenge for taxation when AI agents execute economic decisions based on hidden rules. This creates 'double unobservability' – both productive ability and the AI's preference-to-execution mapping are unobserved – leading to situations where identical tax-base responses can have different welfare outcomes. The study proposes an optimal-tax condition that incorporates a response-weighted execution wedge to address this complexity, suggesting potential corrective benefits or additional costs for higher marginal tax rates depending on the AI's execution behavior.

Why it matters

The emergence of AI agents in economic decision-making fundamentally alters the landscape for income taxation, introducing new complexities for revenue collection and welfare optimization. Understanding and adapting tax policy to account for 'welfare-opaque income' is crucial for maintaining effective fiscal policy and ensuring equitable outcomes in an AI-driven economy.

Key insights

• AI agents implementing economic choices via hidden rules introduce a 'double unobservability' problem for tax authorities, as both productive ability and the AI's decision-making logic are unobserved.
• The income generated under these conditions is termed 'welfare-opaque' because the same observable tax-base response can lead to different welfare consequences.
• Tax-base statistics can be identical even when the welfare effects of tax reforms differ, challenging traditional taxation models.
• The research derives an optimal-tax condition that modifies familiar sufficient statistics by adding a 'response-weighted execution wedge'.
• This new condition implies that a higher marginal tax rate could offer corrective benefits under 'local over-execution' by the AI, but also incur additional costs.

Source

arXiv — Computers and Society — https://arxiv.org/abs/2609.20425