Analysis: original interpretation of cited primary evidence.
Deutsche Bank and Ant Digital Technologies published a proposal on November 9 for using AI agents in large corporate payments. Their framework separates the software that reasons about a payment, the rules that authorize it, and the systems that actually move the money. That separation is the strongest part of the proposal: an agent can help choose a route without being able to enlarge its own authority.
Consider a company moving money across borders. An agent could weigh fees, exchange rates, settlement timing and available routes. A separate system would still decide whether the proposed payment falls within the company’s permitted amounts, destinations and approval rules. If the agent makes a bad recommendation, the payment can be stopped before settlement. This matters whether the final transfer uses a bank rail or a digital settlement system.
The authors also propose giving agents verifiable identities tied to accountable owners, with permissions that can be revoked. That makes an incident easier to investigate: a treasury team needs to know which agent acted, who delegated authority and which rule let the payment through. The announcement describes Ant Digital’s Anvita Flow as a case study. It does not disclose a production rollout at Deutsche Bank, customer payment volumes or evidence that the controls have been independently tested at scale.
The practical test for an institution is whether it can stop or reverse an agent’s permission before the next transfer, and reconstruct how an earlier payment was authorized. Faster routing is useful only if those controls remain enforceable when the agent makes a mistake.


Join the conversation
Share your perspective and reply to other readers in Community.
Discuss this article →