Particle.news

Banks Put Customers on the Hook for AI Agent Account Activity

Shifting legal responsibility to customers, the move steers third‑party tools toward read‑only, regulated connections.

Overview

  • Citibank updated its Online User Agreement on Wednesday, Nov. 18, 2026, saying customers who give AI agents access are responsible for any actions those agents take, including harm from leaked credentials or hacks.
  • JPMorgan Chase, Bank of America, Wells Fargo and other large banks have changed terms this fall to restrict credential sharing, treat agent actions as customer‑authorized, and favor read‑only access through vetted open‑banking connectors.
  • Banks have begun actively blocking or limiting unvetted third‑party agents and are rolling out technical controls to detect, shut down, or downgrade risky agent connections to read‑only modes.
  • AI vendors are responding with safer options, for example Anthropic’s read‑only personal finance tool, while regulators in the U.S. and Canada consult with banks and industry teams to investigate risks and harden systems.
  • For customers this means more responsibility for their account security and fewer apps that can move money on their behalf, a shift that could preserve bank control over payment channels and slow third‑party competition.