The International Monetary Fund (IMF) has released a note How Agentic AI Will Reshape Payments, on Agentic AI in Payments, examining how artificial intelligence may move from supporting financial decisions to directly initiating and executing them.
Unlike traditional AI, "Agentic AI" moves beyond analysis to transaction initiation, effectively acting as a proxy for human decision-making. To manage the inherent risks of AI’s probabilistic (predictive) nature, the IMF proposes a Three-Layer Conceptual Framework:
Layer 1 for Intent and Orchestration (AI reasoning),
Layer 2 for Deterministic Control and Authorization (rules-based checks),
Layer 3 for Settlement (legal finality).
This separation ensures that while AI can propose actions, the execution remains auditable, rules-bound, and predictable.
The shift toward agentic infrastructure is already visible through initiatives like OpenAI/Stripe’s agentic commerce and Visa’s agent registration frameworks. While these systems offer lower operational costs and enhanced liquidity management, they introduce complex challenges regarding liability ambiguity and systemic risks from correlated agent behaviour.
The IMF emphasises that current regulatory frameworks must evolve from "Know-Your-Customer" (KYC) to "Know-Your-Agent" (KYA) protocols. In this evolving landscape, India serves as a primary benchmark; its Unified Payment Interface (UPI) already integrates payments, data, and intelligence as a unified stack, utilizing embedded AI for real-time fraud detection and automated reconciliation to support its global Digital Public Infrastructure (DPI) exports.
Key Architectural and Risk Benchmarks
The Three-Layer Model: Distinct separation between AI "reasoning" and the "deterministic" execution of payments to preserve system trust.
Economic Value: Massive reduction in "human latency" and search costs, potentially unlocking trillions in trapped liquidity and operational efficiency.
Shift to KYA: A mandate for regulators to develop Know-Your-Agent frameworks to track authorization traceability and liability.
Liability Ambiguity: A critical risk factor where it remains unclear who is responsible (the user, the developer, or the agent) for an autonomous AI "hallucination" leading to a payment.
Correlated Risks: The danger of "herding" behavior where millions of agents using the same underlying LLM make identical market moves simultaneously, triggering flash crashes.
Operational Strategy: Implementation of "Kill Switches" and interruption mechanisms to halt autonomous agents during emergency market volatility.
What is "Agentic AI"?
Agentic AI refers to AI systems that don't just answer questions but can actually perform tasks autonomously to achieve a goal. In a standard payment, a human must click "Pay." With Agentic AI, a user might give a goal like "Find the best price for a flight to Munich and book it." The AI "Agent" then navigates websites, compares prices, chooses the flight, and initiates the payment on its own. The IMF 2026 Note explains that while this is highly efficient, it requires new "Control Layers" to ensure the AI doesn't accidentally spend more than the user intended or fall victim to a digital "trap" set by hackers.
Policy Relevance
Scaling UPI’s Global Dominance: By embedding AI for fraud detection and alternative credit scoring directly into the stack, UPI aligns with the IMF’s vision of a "Unified Intelligence Stack," making it the ideal DPI exportfor nations seeking agent-ready payment rails.
Future-Proofing the RBI Regulatory Sandbox: The IMF’s KYA (Know-Your-Agent) recommendation provides a blueprint for the Reserve Bank of India to test "Autonomous Agent Mandates" where users can set cryptographically verifiable limits for their AI assistants.
Protecting Against "Correlated AI" Shocks: The warning on correlated agent behaviour is critical for SEBI and RBI to prevent systemic instability if Indian fintechs all adopt similar AI models for automated high-frequency retail trading.
Closing Financial Inclusion Gaps: Agentic AI can act as a "financial navigator" for low-literacy users in rural India, using voice-based agents to handle complex UPI/Bharat Bill Pay workflows, provided Layer 2 deterministic controls are in place.
Cross-Border Orchestration: As India links UPI with Singapore, the UAE, and France, the IMF’s framework helps standardise how AI agents "negotiate" currency exchange rates and transaction paths in real-time.
Follow The Full Note Here: IMF Note 2026/004: How Agentic AI Will Reshape Payments