Deutsche Bank and Ant Digital Technologies publish joint paper on AI agents in wholesale finance
The November 9, 2026 paper lays out where AI agents could act in treasury, cross-border payments and trade finance, and proposes limits on their authority.
Also in: 日本語 · エージェント経済 · 한국어 · 에이전트포스트 · 简体中文 · 链智日报
The joint paper sets out where AI agents could act in treasury, cross-border payments and trade finance, and proposes limits on their authority.
Deutsche Bank and Ant Digital Technologies published a joint white paper on November 9, 2026, examining how AI agents could be used in institutional finance and what controls would be needed to deploy them, the companies said. The paper, “Agentic Payments in Wholesale Finance,” is available for download from both companies’ websites.
The paper identifies three areas where agents could take on work. In treasury and liquidity management, it says agents could forecast shortfalls and surpluses, assess available response options and carry out compliant transfers, investments, external financing and foreign exchange operations. In cross-border payment orchestration, they could weigh foreign exchange rates, payment rails and execution timing against cost, speed, availability, regulatory compliance and settlement risk. In trade finance, they could read trade documents, flag data discrepancies, coordinate exception handling and prepare payment or financing instructions once conditions are met and approvals obtained.
The paper argues for “bounded agentic execution” rather than unrestricted autonomy. It draws on a three-layer reference architecture proposed by the International Monetary Fund: intent and orchestration, which covers agent reasoning and workflow; control and authorisation, which applies rule-based constraints and keeps actions inside set limits; and settlement, which carries out the final transaction. The design keeps AI-driven reasoning separate from transaction execution.
The safeguards it sets out include authorisation limits that are explicit, bounded and revocable, binding of agent identity, and Know Your Agent (KYA) verification. The paper presents Anvita Flow, an agentic platform developed by Ant Digital Technologies, as a case study in orchestrating payment workflows while keeping controls on agent authority.
In a foreword, Zhuoqun Bian, president of Ant Digital Technologies, wrote that agents need verifiable identity bound to an accountable owner, and that their authority must be explicit, machine-readable, bounded and revocable. Bian also wrote that decision-making, control and execution must be separated so an agent may propose but cannot extend its own authority.
Sabih Behzad, head of digital assets and currencies transformation at Deutsche Bank, wrote in a separate foreword that the value lies not simply in executing payments automatically but in improving how decisions are made, coordinated and controlled.