AI Models Exhibit Gender-Based Risk Tolerance in Financial Decisions, Study Finds

Large language models from leading tech firms systematically alter their financial risk behavior when prompted with gender-specific identities. Research reveals AI systems become markedly risk-averse when assuming female personas, mirroring real-world statistical trends where women demonstrate greater financial caution.

The study employed OpenAI, Google's Gemini 2.0 Flash-Lite, DeepSeek Reasoner, and Meta's models in controlled experiments using the Holt-Laury task—a standard economic assessment measuring risk tolerance through progressive lottery choices. Models prompted as female consistently delayed switching from SAFE to risky options compared to male-prompted counterparts.

DeepSeek and Gemini exhibited the most pronounced effects, with risk profiles diverging by up to 40% between gender conditions. This behavioral mirroring raises questions about the datasets used to train financial decision algorithms in cryptocurrency markets, where risk appetite directly impacts trading strategies.