Thank you, Graeme, and thank you SWIFT for the kind invitation.
The theme of this year’s conference is digital finance for AI-driven economies. This is an important topic for the IMF because an efficient and secure payments system is critical to economic growth and stability.
Clearly, as the theme suggests, the financial system, and payments in particular, will need to evolve. If AI is used as widely as proponents expect, a great deal of economic activity will be undertaken by AI agents rather than by people. The current payments system was not built for autonomous agents operating continuously and at scale.
At the same time, AI has extraordinary potential to transform the payments system we already have, even before we get to the more difficult questions about how the financial system itself may need to evolve.
Today, I want to explore three points: first, how AI can improve the payments system; second, whether and how the financial system needs to be reengineered to support an AI-intensive world; and third, how policymakers should respond to certain risks that are unique to AI.
I. What AI can do for payments
Let me start with how AI can improve the payments landscape.
One way is by reducing information frictions and switching costs. AI agents can assess fees, services, exchange rates and execution quality continuously, routing each transaction to the most efficient provider. That lowers barriers to entry, increases competition and puts pressure on incumbents to improve their prices and services. This applies not just to financial services, but to all forms of commerce. One might even wonder whether a sustained reduction in information frictions and switching costs could have macroeconomically significant consequences for inflation.
Second, agents can automate complex transactions that today require extensive human involvement. They can manage the mechanics of paying, coordinating machine-to-machine payments and optimizing the timing based on liquidity, costs, and contractual obligations. Some of this is possible today through programmability. But programmability only takes you so far. It requires explicit instructions for each task and operates within predefined parameters. By contrast, agentic systems can interpret context and operate without requiring a new set of instructions for each transaction. That would open up more commerce to automation, increasing the volume and speed of transactions.
Third, agentic AI could substantially lower the costs of regulatory reporting and compliance by automating the collection, reconciliation and reporting of data across institutions and jurisdictions. Agents could continuously monitor transactions and positions against regulatory requirements, identify gaps in real time, and produce compliance documentation.
The benefits of AI become even more powerful in the context of cross-border payments, which bridge different legal systems, currencies and regulatory requirements. Agents could automate activities such as customer due diligence and sanctions screening, improving the speed and consistency of compliance. And the benefits from agent-enabled price comparison are likely to be even larger than in the domestic context, given the incremental frictions faced by retail consumers making cross-border payments.
I would not underestimate the scale of the benefits of applying AI to the existing payments architecture. Payments are about much more than just the public infrastructure that supports them. A full payments product requires regulatory compliance, dispute resolution, and customer support. Those functions are needed regardless of whether payments run on blockchain rails or other infrastructures.
Many of the benefits promised by a reimagined financial system can instead be delivered by upgrading the core payment systems that we already have. That is where AI can help. In the short run, this is where we are likely to see more tangible benefits.
The scale alone makes the case for investment. IMF staff estimate that cross-border payments move nearly $1 quadrillion a year. At that scale, even improvements measured in basis points can translate into very large gains.
II. How the financial system might evolve
As AI agents become a significant part of economic activity, the financial system will need to evolve alongside them. It must become more digital, more interoperable and more programmable.
A range of payment technologies could supply the infrastructure needed for autonomous agents to transact across financial networks. Tokenization may be especially promising, since tokenized assets are natively programmable and can settle almost instantly, at any hour.
We are already seeing progress. CHIPS, for example, is leading a multi‑bank initiative to create a shared tokenized deposit settlement platform.
The challenge for policymakers is to support innovation without compromising stability. That begins with legal and regulatory frameworks that permit different payment technologies to compete on their merits.
The law needs to clearly establish the legal nature of tokens, including the rights and claims of purported owners and parties with varying degrees of contractual privity. It must also define the relationship between tokenized assets and their real-world equivalents. Authorities should avoid erecting unnecessary obstacles to interoperability across networks. And liquidity frameworks will need to keep pace because while atomic settlement removes counterparty risk, it also raises intraday liquidity needs by requiring prefunding.
If banks continue to embrace tokenization, it naturally raises the question of how central banks should respond, as the existing bank-based payments architecture rests on a foundation of central bank money that is not currently tokenized.
Stablecoins, too, raise questions of the form of liabilities issued by the government. To maintain a stable value and function as a payment instrument, they need to be backed by safe and liquid assets like short-term government obligations, central bank deposits, or insured commercial bank deposits that are ultimately supported by central bank money.
At the Fed’s Jackson Hole conference this year, we had very interesting discussions about whether central banks should offer new forms of central bank money, such as tokenized reserves, and whether they should extend access to central bank money to new parties. Finance ministries, too, might consider whether they should issue government obligations in tokenized form.
We are seeing a wave of experimentation across these domains. For example, the ECB’s recently launched Project Pontes is designed to enable wholesale transactions on distributed ledger platforms to settle in central bank money. The Bank of England is requiring certain issuers of stablecoins to hold central bank deposits.
The diversity of these approaches reflects varying country circumstances and underscores why policymakers should keep their options open, observe how developments unfold, and consider tailored solutions as the market matures. This is an area where the IMF supports central banks in evaluating alternative pathways through technical assistance.
It is certainly appropriate for central banks to provide new public infrastructure that supports competition, such as tokenized reserves. But these services should be constructed in a way that does not lock out new technologies and business models that might emerge from market-based competition. The objective should be to create an enabling environment, while ensuring the integrity and stability of the payment system.
Preserving space for competition is not just about increasing efficiency and innovation. It is also a financial stability issue. Policymakers that encourage overly narrow solutions often discourage heterogeneity in business models that can be a source of resilience. When the inevitable shock comes, similar institutions tend to fail in the same way, creating systemic risk.
III. Approaching AI risk
The development and diffusion of AI technologies may pose broader risks that need to be considered. Cyber risks are of course at the top of the list. AI is reshaping cyber risk by accelerating the speed, frequency, and breadth of vulnerability discovery and potential exploitation. Tokenized markets depend heavily on shared, complex infrastructure and a small number of third-party providers. That means a cyber incident can propagate widely enough to become systemic.
Fortunately, we have not yet seen a macro-critical cyber event. Private risk management and the existing supervisory architecture have contained risks. That approach still applies: strong governance, technical controls that limit the blast radius of a breach, and robust incident response and recovery. We need more coordination among standard setters and supervisory networks, and between governments and industry.
As frontier AI tools are deployed by cyber actors, defenders will need those same tools to respond. And that gets to a broader debate about whether increasingly powerful AI systems could pose such severe risks to society as a whole that new regulation is needed to slow the pace of development.
These concerns deserve to be taken seriously. The IMF contributes to this vital debate by bringing a macro-financial lens to questions of regulation and risk. We focus on where market failures may arise; where risks can become systemic or macro-critical; and where private incentives may diverge from the public interest.
In AI, that may include major externalities from defective products, information gaps between developers, users, and regulators, or operational dependencies on a small number of providers. It could also include the loss of competitiveness and concentration of market power that may result from the imposition of a poorly designed regulatory framework that creates barriers to entry.
The starting point should be to identify demonstrable market failures and evaluate whether existing regulatory frameworks are sufficient to address them. To the extent that new interventions are needed, they should be proportionate, targeted, and internationally coordinated, while taking great care not to discourage competition and innovation.
AI could be transformative for the payments system. Handled well, AI can help build a payments system that is more efficient, more competitive, and better suited to the economy of the future.
For policymakers, the challenge is to enable this transformation, while safeguarding against risks. The IMF supports our membership by helping them assess spillovers and vulnerabilities from this transformation.
More broadly, the Fund also facilitates the international dialogue needed for the payments system to remain open and stable—consistent with our 80-year mandate to ensure the stability of the international monetary and financial system.
Thank you very much.
Compliments of the International Monetary FundThe post IMF | AI and the Future of Payments Policy – Remarks by Dan Katz, First Deputy Managing Director at Sibos 2026 Conference first appeared on European American Chamber of Commerce New York [EACCNY] | Your Partner for Transatlantic Business Resources.
