AI speeds up fraud

Fraud used to look like a clumsy fake email. In 2026, it can look like your bank’s real login page, sound like your daughter on the phone, and arrive seconds after a real transaction because generative artificial intelligence can produce convincing text, images, audio, and code on demand. (bankinfosecurity.com) (consumer.ftc.gov) That speed is the story. Intuit security architect Tony Gauda said on April 9 that artificial intelligence is accelerating fraud “at unprecedented speeds,” which means companies have less time to decide whether a login, payment, or account change is real before the customer notices any delay. (bankinfosecurity.com) Old fraud systems were built like airport security with a short checklist. If a payment was over a set dollar amount or came from a new device or country, a rule fired; if it did not match the checklist, it often passed. (bankinfosecurity.com) That worked better when fraud patterns changed slowly. It works worse when one scammer can use artificial intelligence to generate thousands of fresh messages, voices, and account personas that all look slightly different, because fixed rules are easiest to beat once criminals know where the line is. (fbi.gov) (ftc.gov) One abuse method getting more attention is voice cloning. The Federal Trade Commission warned in April 2024 that scammers can copy a family member’s or boss’s voice and use the fake call to push for money or account information right away. (consumer.ftc.gov) Another is impersonation at scale. The Identity Theft Resource Center said reported impersonation scams rose 148 percent in its 2025 Trends in Identity Report, with criminals most often pretending to be a general business or a financial institution. (idtheftcenter.org) The losses are no longer niche. The Federal Trade Commission said consumers reported more than $12 billion lost to fraud in 2024, and the Federal Bureau of Investigation said cyber-enabled crime defrauded Americans of nearly $21 billion in 2025, with cryptocurrency and artificial-intelligence-related complaints among the costliest. (ftc.gov) (fbi.gov) So the engineering problem has changed from “block bad transactions” to “score trust in real time.” Gauda’s point is that companies now have to judge identity, device history, behavior, and context in the moment, often before a checkout, password reset, or tax refund flow is finished loading. (bankinfosecurity.com) That creates a second problem: friction. If every unusual login triggers a text code, selfie check, or support call, good customers get punished, carts get abandoned, and fraud teams end up training users to hate security. (bankinfosecurity.com) The new model is quieter. Instead of stopping everyone at the door, companies try to watch for behavior that does not fit the person behind the account, like typing rhythm, device reputation, session timing, or a sudden shift from normal patterns, and then step up checks only on the riskiest cases. (bankinfosecurity.com) That means fraud defense is starting to look less like a lock and more like a live credit score for trust. The score has to update in milliseconds because the same artificial intelligence tools that help companies automate service also help criminals test, revise, and relaunch scams almost instantly. (bankinfosecurity.com) (fbi.gov) The uncomfortable part is that there is no finish line. The Federal Trade Commission said there is no single fix for artificial-intelligence-enabled voice cloning, and Gauda’s interview points to the same conclusion for fraud more broadly: faster attacks are turning trust, identity, and anomaly detection into always-on infrastructure instead of a back-office checklist. (ftc.gov) (bankinfosecurity.com)