Deutsche Bank and Ant Digital Technologies are exploring how AI agents could take on a greater role in wholesale payments.
The companies have published a joint white paper titled Agentic Payments in Wholesale Finance.
It examines how AI agents could help businesses and financial institutions manage treasury operations, cross-border payments and trade finance.
Rather than simply following fixed instructions, AI agents could assess information, compare options and initiate transactions within limits set by financial institutions.
Where AI Agents Could Be Used
Treasury teams could use the technology to monitor cash positions, spot potential funding shortfalls and assess whether to move funds or seek financing.
Agents could also help with cross-border payments by comparing exchange rates, fees and payment networks.
They would assess the available options against costs, speed and regulatory requirements.
Trade finance is another potential use case. Agents could review documents, flag discrepancies and coordinate checks before financing or payments are approved.
The report suggests these applications could reduce manual work and improve efficiency, although the benefits still need to be proven in practice.
Ant Digital Technologies’ Anvita Flow platform is included as an example of how agent-based payment workflows could be designed.
Keeping AI Agents Under Control
The report cautions against giving AI agents unrestricted control over financial transactions.
It draws on a framework proposed by the International Monetary Fund (IMF) that separates AI decision-making from payment authorisation and settlement.
Independent systems would check transactions against institutional rules and compliance requirements.
Human approval would still be required for high-value, high-risk or exceptional cases.
The paper calls for verified agent identities, clear limits on what agents can do and the ability to revoke their permissions.
Financial institutions would also need to keep records of decisions and transactions for audit purposes.
It recommends starting with narrowly defined applications and giving AI agents more responsibility only after demonstrating reliable performance and effective controls.
Featured image: Edited by Fintech News Hong Kong, based on image by Ant Digital Technologies
