AI will answer your money question. It won't read the person asking.

The first stop for a money question is increasingly a chatbot, and the first stop now answers it without reading the person asking. A University of Canberra study by Bomikazi Zeka and Raechel Johns ran ChatGPT, Claude, and Perplexity through five scenarios built around real human strain: a 22-year-old saving for a home deposit in a cost-of-living squeeze, a pregnant woman planning maternity leave with a partner who does not share money, a modest-income single parent pushed toward cryptocurrency by a relative. The researchers stripped identifying attributes and ran each scenario five times in fresh sessions. They handed the models a vulnerability in plain language and watched the polished reply come back anyway. The pattern is vulnerability-blindness: structured advice that never registers the constraint it was told about. The same reply that would help a financially stable planner with no dependents walks straight past a person on the edge. In places, the models recommended moves that would deepen the harm the prompt was built around. Zeka and Johns close with a constructive bound, captured in their study: AI is a powerful tool for financial fact-finding and brainstorming. The test for a reader is simple. If the answer reads like a generic financial plan and never names the person in the prompt, the model is in vulnerability-blind mode. Stop, cross-check, escalate.