Vals AI and NextLadder Ventures released a benchmark on Nov 13, 2026 that tested 28 general‑purpose AI models in 256 simulated low‑income financial counseling scenarios. The models frequently gave advice before collecting basic facts like income or household size. Even the strongest models left out about one‑fifth of key client information, and in one out of ten chats they relied on wrong assumptions. Free versions of ChatGPT and Gemini missed critical points in roughly half of crisis conversations.
Why it matters
Low‑income families could receive incomplete or harmful financial guidance from popular chatbots.
