Project Agenti, a collaboration between the BIS Innovation Hub Swiss and Toronto Centres, the Bank of Canada and the Swiss National Bank, is exploring the feasibility and policy value of using agentic AI to model payment behaviour and intraday liquidity dynamics in wholesale payment systems.
By examining how AI-driven behaviour affects payment timing, liquidity usage, gridlock formation and the propagation of shocks, the project aims to help central banks identify emerging risks, strengthen stress testing and inform future policy, oversight and payment-system design.
Why it matters
Wholesale payment systems are becoming faster, more automated and increasingly data-driven. As AI becomes more capable of making autonomous decisions, it could influence how financial institutions manage payments and intraday liquidity, and potentially introduce new sources of systemic risk.
Project Agenti gives central banks a controlled, auditable and risk-free environment to explore these developments before such technologies are deployed at scale. This strengthens the analytical tools available for stress testing, scenario analysis and payment-system design, while supporting responsible innovation in wholesale payments.
The project
At the core of Project Agenti is an Agentic AI Payment Simulator, which models an RTGS-like environment. AI agents represent banks and other payment-system participants, each with its own objectives, constraints and strategies for managing payments and liquidity.
The project builds on early research showing that AI agents can interpret and implement structured payment and liquidity policies, such as prioritising payments, maintaining liquidity buffers and sequencing transactions coherently. The simulator will be used to test whether AI agents can realistically reproduce the behaviour of existing participants and to explore how different behaviours and degrees of autonomy could affect liquidity demand, payment delays, gridlock and system resilience.