Author(s)

Xiuling Zhou, Shuang Diao

Corresponding Author

Shuang Diao

Abstract

This paper proposes a governance-ready framework for identifying and mitigating decision biases introduced by generative AI in robo-advisory, integrating explainability constraints, causal robustness, and compliance executability within a layered architecture spanning data, features, forecasting, optimization, and risk governance. Counterfactual evaluation and constrained decision optimization coordinate return, risk, and compliance objectives, and a semi-synthetic, empirically calibrated study indicates that, at equal risk budgets, generative-AI-enhanced advisory improves allocation diversity, communication personalization, and scenario responsiveness while remaining susceptible to prompt injection, hallucination, overconfidence, and interaction-induced risk-preference drift. An integrated protection bundle composed of decision evidence cards, policy corridors, robust optimization, and human-in-the-loop governance reduces erroneous extrapolation and strategy whipsawing, enhances suitability alignment, and strengthens tail-risk control under stress, offering a pragmatic deployment roadmap.

Keywords

Generative AI; Robo-advisory; Decision bias; Causal robustness; Investor protection; Explainability; Reinforcement learning; Robust optimization; Suitability; Compliance