Banks Face New Trust and Security Challenges as AI Agents Handle Transactions

· Original article ↗

Summary

Banks are deploying internal AI agents as customers gain agents that can transact for them. The article examines authentication, fraud, authority, liability and oversight risks, citing early deployments and a test payment.

Key points

  • LHV Bank has launched agentic AI for customer emails after a proof of concept, emphasizing guardrails, explainability, auditability and accountability.
  • BNY describes its AI agents as digital employees with identities, access controls and defined workflows.
  • Danske Bank and Mastercard demonstrated an AI agent booking and paying for a coffee tasting with a bank-issued card.
  • Federal Reserve Board member Christopher Waller distinguished agent-assisted purchases from delegated agents authorized to shop and pay; delegation raises greater risk of unintended purchases.
  • Banks may need to verify an agent's authority, adapt fraud detection calibrated to human behavior, and establish who is liable when an agent makes a mistake.
  • A Mastercard executive recounted an unnamed bank receiving a call from an AI agent seeking to renegotiate a customer's credit card terms; the bank had no procedure for handling it.

Article Details

Topic
Banks’ deployment of internal AI agents and customer-facing agentic payments, with emphasis on trust, authentication, authority, fraud, liability, and controls.

People

Vendors

Products

Countries