Study Finds 3 AI Models Miss Vulnerability in Financial Advice, Sometimes Worsening Risk
Updated
Updated · The Conversation · Aug 16
Study Finds 3 AI Models Miss Vulnerability in Financial Advice, Sometimes Worsening Risk
3 articles · Updated · The Conversation · Aug 16
Summary
Five test scenarios run five times each found ChatGPT, Claude and Perplexity gave practical financial answers but often failed to recognize when users were vulnerable and could even steer them toward harmful choices.
The blind spots appeared even when prompts explicitly described risk factors such as a cost-of-living squeeze, separate household finances during pregnancy, or a single parent with modest income considering cryptocurrency.
ChatGPT gave the most detailed guidance but often missed embedded vulnerability; Perplexity was the most cautious and most likely to suggest professional help, while Claude offered broad recommendations that leaned toward self-directed planning.
Researchers said the failures reflect AI’s limits in reading human context and its tendency to reproduce social assumptions, underscoring that financial advice still requires judgment about values, stress, family dynamics and risk tolerance.
The study argues AI can help with fact-finding and brainstorming for financially literate users, but regulators need safeguards and transparency rules on AI-generated advice and sensitive consumer data handling.