Behavioral AI moves into fraud defence
Reports this week describe behavioural biometrics—how users type, swipe and interact on mobile—as an active tool for spotting banking fraud. Media outlets highlighted prototype deployments and vendor claims, and public agencies are beginning to use AI decisioning for audits with mixed results, raising governance and explainability questions (thedailystar.net) (postandcourier.com).
Banks and tax agencies are starting to use “behavioral” artificial intelligence that watches how people type, swipe and move, not just what password they enter. (thedailystar.net) The idea is simple: a phone user’s rhythm can work like a fingerprint. The Daily Star reported on April 11 that researchers in Bangladesh are testing systems that compare typing, swiping and scrolling patterns with a customer’s usual location, timing and transfer size. (thedailystar.net) That approach aims at a specific weakness in mobile banking: many scams now use valid credentials. If a victim is tricked into sharing a one-time password or login code, a rules-based system may see a normal sign-in, while a behavior model may flag that the person holding the phone is acting differently. (thedailystar.net) Bangladesh is a useful test case because mobile money is already huge there. Bangladesh Bank data cited by The Daily Star put registered mobile financial service users at more than 144 million in January 2026, including about 570,000 youth accounts. (thedailystar.net) Fraud pressure is already visible in enforcement numbers. On May 9, 2024, The Daily Star reported that the Bangladesh Financial Intelligence Unit had suspended 48,586 personal mobile financial service accounts over suspected links to online gambling, betting and hundi. (thedailystar.net) The same pattern is moving into government decision-making in the United States. The Post and Courier reported on April 12 that the South Carolina Department of Revenue will begin using artificial intelligence in 2026 to help choose which business taxpayers to audit. (postandcourier.com) South Carolina’s own audit page says the agency already reviews filed returns alongside information from the Internal Revenue Service, other state agencies and other sources, and sends a notice before an audit begins. Most desk audits are resolved within 60 days, according to the department. (dor.sc.gov) The promise in both cases is speed: software can scan far more signals than a human reviewer can. Vendors such as Outseer market behavioral biometrics as a way to catch account takeovers, scams and social-engineering fraud in real time, even when credentials appear legitimate. (outseer.com) The unresolved question is how much of that judgment can be explained after the fact. The National Institute of Standards and Technology says its Artificial Intelligence Risk Management Framework is meant to help organizations build trustworthiness into the design, use and evaluation of artificial intelligence systems. (nist.gov) That leaves banks and agencies with the same test: if a system blocks a payment or helps trigger an audit, the institution still has to defend the decision in plain language. The more these tools move from pilot projects into frontline screening, the harder that explanation becomes. (nist.gov)
