AI Banking Agents Raise New Questions About Financial Control

AI agents can do more than answer questions: they can carry out tasks independently, including financial actions. That capability could make banking more convenient, but cybersecurity experts warn that an agent may act in ways its creators did not foresee, even without malicious intent or outside tampering.

Jake Williams, a faculty analyst at IANS, said autonomy is what defines an AI agent and also creates risk. A system might carry out an assigned task in an unexpected way, leaving a gap between what its user intended and what it actually does.

The stakes rise when an agent has access to a bank account. T.J. Marlin, CEO of Guardrail Technologies, raised the possibility of an agent transferring $100,000 to an unintended recipient after being given account credentials. He said financial agents need clear limits, a way for users to stop them and a record of their actions.

Banks are also using AI to counter potential risks from AI agents. Jonathan Gould, who heads the Office of the Comptroller of the Currency, noted that lenders have spent months bolstering protections and deploying AI as part of that work.

As these tools become more capable, consumers face a choice about how much authority to give them over their money. Experts point to safeguards such as limits on what agents can do, a way to revoke access and records of completed actions.