On November 9, 2026, Deutsche Bank and Ant Digital Technologies jointly issued a white paper titled Agentic Payments in Wholesale Finance, outlining a framework for deploying autonomous AI agents in treasury, cross‑border payments, and trade finance while embedding robust governance controls.

What the White Paper Proposes

The document positions agentic AI as more than a transactional tool; it argues that AI agents can enhance the quality of financial decisions by autonomously evaluating alternatives and proposing actions that align with institutional policy. Rather than granting unrestricted autonomy, the authors advocate a “bounded agentic execution” model that keeps reasoning and execution separate.

Key Architectural Pillars

• Intent and Orchestration – The top layer manages adaptive reasoning, workflow orchestration, and decision proposals.
• Control and Authorisation – A rule‑based engine enforces deterministic constraints, ensuring that any proposed action remains within pre‑approved limits.
• Settlement – The final layer executes the transaction, guaranteeing that only vetted, authorised instructions reach the payment rails.

Applications in Treasury and Liquidity Management

AI agents can forecast liquidity gaps, evaluate funding sources, and propose internal transfers or external borrowing in real time. By automating the assessment of multiple instruments—cash, securities, and FX—the system reduces manual spreadsheet work and speeds up decision cycles. The white paper cites a pilot where a bank’s AI agent identified a liquidity shortfall within minutes of market opening, triggering an automated internal transfer that avoided a costly market‑timed trade.

Optimising Cross‑Border Payments

In cross‑border scenarios, agents weigh currency rates, payment rails, settlement windows, and regulatory constraints to determine the optimal routing. The framework supports multi‑currency execution and can flag potential compliance gaps before a transfer is initiated. This reduces both cost and settlement risk, especially in regions with fragmented payment infrastructures.

Trade Finance Automation

Trade finance documents are notoriously complex. AI agents can parse invoices, letters of credit, and shipping documents to detect anomalies, trigger exception workflows, and generate payment or financing instructions once all conditions are met and approvals are in place. The paper notes that such automation can cut processing times from days to hours, freeing compliance teams to focus on higher‑value tasks.

Governance and Risk Safeguards

The authors stress the importance of explicit, machine‑readable authorisation limits. Agents must be bound to a verifiable identity and an accountable owner, with the ability to revoke authority at any time. The “Know Your Agent” (KYA) protocol mirrors KYC practices but for AI agents, ensuring that each decision chain is traceable and auditable. The control layer’s deterministic rules act as a safety net, preventing any drift beyond agreed boundaries.

Case Study: Anvita Flow

The white paper showcases Anvita Flow, Ant Digital Technologies’ agentic platform, as a real‑world implementation. Anvita Flow demonstrates how institutions can orchestrate complex payment workflows while maintaining strict controls over agent authority. The platform’s architecture aligns with the IMF’s three‑layer reference model, and its deployment in a regional bank’s treasury department reportedly reduced manual intervention by 35%.

Implications for the Industry

By formalising the separation between reasoning and execution, the framework addresses a key barrier to AI adoption in finance: trust. Institutions that can prove their agents act within well‑defined rules are more likely to gain regulatory approval and stakeholder confidence. Moreover, the modular architecture allows banks to incrementally introduce agentic capabilities—starting with routine treasury checks before moving to full cross‑border orchestration.

Competitive Landscape

While several fintechs offer AI‑enhanced analytics, few provide a full governance stack that couples autonomous reasoning with enforceable execution controls. Existing solutions like Bloomberg’s AI‑powered analytics or Visa’s AI fraud detection focus on data insights rather than transactional autonomy. The proposed model therefore fills a niche for institutions seeking end‑to‑end automation without compromising oversight.

Adoption Challenges

Implementing agentic payments requires significant investment in identity management, rule‑engine development, and audit logging. Regulatory bodies will scrutinise the agent’s decision logic, demanding transparency and explainability. Moreover, legacy core banking systems may need integration adapters to expose the necessary APIs, adding complexity to the rollout.

Business Opportunities

Financial institutions that adopt the bounded agentic model can unlock efficiencies across their treasury, payments, and trade finance operations. By reducing manual processing and accelerating decision cycles, banks can lower operating costs and improve customer service. Fintechs that build modular agentic layers could position themselves as trusted partners for banks looking to modernise their payment infrastructures.

Market Landscape

The global AI adoption rate in banking is projected to reach 70% by 2025, according to McKinsey’s 2023 report on AI in finance. Cross‑border payment volumes are expected to grow 12% annually, driven by e‑commerce expansion and regulatory harmonisation. Yet, only 15% of banks have implemented AI agents that can autonomously propose and execute payments, indicating a significant gap between capability and deployment. The agentic payments framework aligns with the IMF’s push for safer, faster, and more transparent payment systems, and it dovetails with open banking initiatives that expose core banking data to third‑party developers.

Top Insights

• Agentic payments separate reasoning from execution, mitigating risk while preserving autonomy.
• Governance hinges on machine‑readable authorisation limits and a robust Know Your Agent protocol.
• Early pilots show a 35% reduction in manual treasury tasks with platforms like Anvita Flow.
• Regulatory scrutiny focuses on explainability, traceability, and revocability of agent authority.
• Adoption barriers include legacy system integration and the need for dedicated AI governance teams.

Get in touch with our fintech expert