Deutsche Bank and Ant Digital Technologies release white paper on agentic payments

From insights to actions

Artificial intelligence is moving beyond just generating insights to creating systems that can interpret goals and execute actions independently. Deutsche Bank and Ant Digital Technologies released a white paper exploring how these 'agentic' systems can function within wholesale finance. The goal is to move from simple automation to agents that can evaluate alternatives and make decisions.

Wholesale financial institutions are currently grappling with how much authority to delegate to these autonomous systems. The report addresses the core tension between operational efficiency and the need for secure, auditable decision-making. It seeks to provide a roadmap for integrating these agents into core functions like treasury management and trade finance.

As PR Newswire reported, the white paper aims to establish governance frameworks for responsible deployment. The institutions want to ensure that as AI becomes more capable of carrying out complex operations, the human oversight remains intact. This transition requires a shift in how banks think about the role of software in high-stakes financial environments.

Key areas for automation

The report identifies three specific areas where agentic payments can provide significant strategic value. In treasury and liquidity management, AI agents can forecast shortfalls and surpluses. They can then assess response options and execute compliant actions across investments, foreign exchange operations, and internal fund transfers.

Cross-border payment orchestration is another primary focus for the new technology. Agents can analyze multiple variables simultaneously, such as exchange rates, payment rails, and execution timing. This allows for the optimization of transaction costs, speed, and settlement risk while maintaining regulatory compliance. These tasks are often too complex for simple rule-based scripts to handle efficiently.

Finally, the white paper looks at trade finance, where agents can interpret complex documentation and identify data discrepancies. Once the necessary conditions are met and human approvals are obtained, the agents can coordinate exception handling and generate payment instructions. This helps streamline the management of complex transactions that involve many moving parts and manual checks.

The bounded execution model

A central theme of the report is the concept of 'bounded agentic execution' rather than unrestricted autonomy. The authors argue that agents must operate within explicit, machine-readable limits that can be revoked by a human owner. This ensures that an agent can propose a course of action but does not have the power to extend its own authority beyond what was originally granted.

To achieve this, the white paper draws on a three-layer reference architecture proposed by the International Monetary Fund (IMF). The first layer handles intent and orchestration, which oversees the reasoning and workflow of the agents. The second layer focuses on control and authorization, enforcing deterministic, rule-based constraints to keep actions within predefined boundaries.

The third layer is dedicated to settlement, which is the actual execution of the financial transaction. By deliberately separating AI-driven reasoning from the final execution, the framework ensures a safety buffer. This structure allows institutions to benefit from AI's processing power while maintaining a hard stop on the final movement of money.

Identity and verification

For these systems to work in a regulated environment, the white paper emphasizes the need for verifiable identities. Every agent must be bound to an accountable owner so that actions can be traced back to a human or corporate entity. This is a fundamental requirement for any institutional finance application involving the movement of large sums of money.

The report introduces 'Know Your Agent' (KYA) verification protocols as a necessary component of this framework. These protocols are designed to verify the identity and permissions of the AI agents interacting with financial systems. This mirrors the 'Know Your Customer' (KYC) standards that are already standard practice in the banking industry.

By establishing these protocols, the institutions aim to create a secure environment for agentic interactions. The white paper argues that these foundations are not optional or purely technical. They are necessary requirements for building trust between the technology, the financial institutions, and the broader regulatory landscape.

Real-world implementation

To move beyond theory, the white paper features a case study on Anvita Flow, a platform developed by Ant Digital Technologies. This platform serves as an example of how financial institutions can orchestrate complex payment workflows in real-world operating environments. It demonstrates how to maintain rigorous controls over agent authority while still allowing for high-level automation.

The platform illustrates the practical application of the three-layer architecture described in the report. It shows how a bank can set up a system where an agent handles the heavy lifting of data interpretation and option assessment. Meanwhile, the final execution remains governed by the strict authorization boundaries defined by the bank's internal policies.

The joint publication serves as a call to action for the wholesale finance industry to begin moving from experimentation to institutional adoption. By providing a clear roadmap and specific safeguards, Deutsche Bank and Ant Digital Technologies are attempting to set the standard for the next generation of financial infrastructure. The full white paper is now available on the official websites of both organizations.