Reserve Management in an AI-Enhanced World

In discussions with central bank clients over the last few years, artificial intelligence (AI) has often been framed in familiar terms: improving productivity, automating workflows or making research more efficient. More recent surveys and official-sector commentary, however, suggest the discussion is starting to shift. Central banks generally remain constructive about AI’s potential, but they’re also proceeding cautiously as financial markets themselves are becoming increasingly AI-enhanced.

According to HSBC’s latest Reserve Management Trends survey, only a relatively small share of reserve managers is currently using AI directly in reserve operations.¹ Most institutions are still in exploratory phases focused on reporting, staff efficiency, portfolio analytics and risk-management support. At the same time, roughly half of respondents believe slow adoption could leave reserve managers at a disadvantage over time, particularly as AI adoption across the broader financial system accelerates.

Survey findings from the independent think tank Official Monetary and Financial Institutions Forum (OMFIF) reinforce that pattern. More than 60% of central banks report that AI still isn’t supporting core operations in a meaningful way.² Current applications remain concentrated in lower-risk areas such as market summarization, report generation, anomaly detection and research support. Interestingly, institutions further along in AI experimentation also tend to be more cautious about its risks.

Examples from the Bank for International Settlements (BIS) illustrate how central banks are proceeding. Bank Indonesia has used machine-learning models to analyze how foreign investor activity affects exchange rates and monetary policy transmission.³ The Reserve Bank of India’s FREE-AI initiative emphasizes governance, ethics and risk assessment before scaling up deployment. Across central banks, AI use remains narrow and tightly controlled for now, with most applications focused on monitoring, surveillance and analytical support rather than core policy or reserve allocation decisions.

Discussions with reserve managers in Latin America suggest a similarly measured approach. In many cases, AI use remains concentrated in areas such as market research, investment analytics, stress testing and streamlining internal committee materials rather than direct portfolio or risk-management decisions. Several institutions also remain reluctant to input portfolio data into AI systems because of concerns around data privacy, cybersecurity and “black box” decision-making. Human oversight remains critical, particularly among lean reserve management teams that have only a few members with both investment and AI expertise.

The Federal Reserve (Fed) has described a similarly cautious approach. Fed Governor Christopher Waller recently noted that strict internal security controls often leave the Fed operating with slightly older technology because protecting sensitive policy information takes priority over rapid adoption. He added that the Fed currently uses AI for functions such as summarizing meetings, but only within tightly controlled guardrails.⁴

That caution is deeply tied to the structure of reserve management itself. Reserve portfolios aren’t designed solely to maximize returns; instead, they’re intended to ensure liquidity preservation, institutional credibility and financial stability. Central banks therefore place a premium on systems that are explainable, auditable, secure and reliable under stress. BIS analysis highlights concerns around AI hallucinations, black-box decision-making, data privacy, third-party dependence and cyber vulnerabilities.

Reserve managers are also facing a growing operational burden. World Bank Reserve Advisory & Management Program (RAMP) surveys point to staffing constraints, modernization pressures and increasing complexity in operational risk management.⁵ The challenge is no longer simply access to information. Reserve managers now face a much larger and more complex flow of information than in the past, while markets themselves are moving faster and becoming harder to interpret. That increases the value of analytical and monitoring tools, but central banks still operate within governance structures that naturally move more cautiously. Many reserve-management processes simply weren’t designed for this level of complexity.

Recent geopolitical developments already illustrate the scale of that challenge. Disruptions tied to the Strait of Hormuz are currently rippling across energy markets, shipping routes, inflation expectations, sovereign spreads, fiscal balances and currency volatility simultaneously. Sanction announcements are also affecting counterparties, payment systems, reserve accessibility and market liquidity across multiple jurisdictions at the same time.

AI is also beginning to shape market behavior itself. The BIS has warned that AI-enhanced trading systems could amplify herding and volatility during periods of stress if market participants rely on similar datasets, signals or model architectures. The IMF has also highlighted rising AI-driven cyber risks, pointing to Anthropic’s Mythos model as an example of how rapidly AI-assisted cyber capabilities are advancing.⁶ Models such as Mythos could dramatically reduce the time and cost needed to identify and exploit vulnerabilities across widely used systems, raising the risk of correlated disruptions across financial institutions, payment networks and shared infrastructure.

For central bank reserve managers, these risks are particularly acute. Reserve portfolios depend on deep and orderly government bond and foreign-exchange markets, especially during periods of stress. A liquidity event made worse by crowded AI-driven positioning or cyber disruptions is exactly the type of scenario reserve managers worry about most.

The evidence so far suggests that central banks will continue adopting AI cautiously, focusing on operational support, surveillance and analytical augmentation, while strategic portfolio decisions will likely remain driven by human judgment and institutional policy frameworks. That caution is understandable. But the brute reality is that markets, information flows and geopolitical shocks are now moving much faster than many reserve-management frameworks were built to handle.

ENDNOTES