Deutsche Bank and Ant Digital Map 'Bounded' AI Agent Framework for Wholesale Payments
Deutsche Bank and Ant Digital Map ‘Bounded’ AI Agent Framework for Wholesale Payments #
Deutsche Bank and Ant Digital Technologies have jointly published a white paper describing how artificial intelligence agents could be embedded into core wholesale banking workflows, with human-defined controls in place throughout.
Released on 9 November, the paper, titled “Agentic Payments in Wholesale Finance,” argues that the question for institutions is no longer whether AI agents can perform complex financial tasks, but how much authority should be delegated to them and under what conditions. Rather than advocating for fully autonomous AI systems, the two firms propose what they call “bounded agentic execution,” a model in which agents can assess data, recommend actions, and coordinate multi-step workflows, but cannot extend their own authority or execute transactions beyond pre-authorised limits.
The paper identifies three wholesale banking use cases it considers most suited for agentic deployment. In treasury and liquidity management, AI agents could continuously monitor cash positions across accounts, currencies, and jurisdictions, anticipate shortfalls or surpluses, and execute approved responses, including internal fund transfers, foreign-exchange trades, and external financing, all within deterministic guardrails. In cross-border payments, agents could dynamically evaluate FX rates, rail availability, compliance requirements, and settlement risk to select an optimal execution path, replacing what is often a fixed operational process with a continuously assessed decision. In trade finance, the paper envisions agents interpreting documentation, flagging discrepancies, coordinating exception handling, and generating payment or financing instructions once required contractual conditions and approvals have been met.
The framework rests on a three-layer architecture the authors describe as separating machine reasoning from final transaction execution. The first layer handles intent and orchestration, the adaptive decision-making of the agent itself. The second enforces deterministic, rule-based constraints that keep any action within pre-defined authorisation boundaries. The third is settlement, where the actual financial transaction is executed only once the layers above have been satisfied.
Both institutions’ leaders contributed forewords stressing the importance of accountability. Sabih Behzad, Deutsche Bank’s Head of Digital Assets and Currencies Transformation, emphasised that the value of agentic systems lies not just in automating payment execution but in improving how decisions are made and controlled. Zhuoqun Bian, President of Ant Digital Technologies, set out conditions she considers non-negotiable: each agent must carry a verifiable identity tied to an accountable owner, its authorisation must be explicit and revocable, and there must be a clear separation between an agent proposing an action and the system granting permission to carry it out.
The paper includes Anvita Flow, an agentic platform developed by Ant Digital Technologies, as a case study in orchestrating complex payment workflows under institutional-grade controls. The full white paper is available on both companies’ websites.
Copy link