The Bank for International Settlements published a 2026 survey by the Irving Fisher Committee on Central Bank Statistics showing that central banks are moving beyond artificial intelligence (AI) experimentation, while oversight, data preparation and provider dependence remain challenges. More than two thirds of reported AI projects were in production, including on a small scale, compared with about half in 2024. Adoption increasingly extends to general purpose productivity tools, although research and statistics continue to account for the most applications. Governance frameworks have expanded, but monitoring remains a weakness. Fewer than 10% of respondents lack AI use guidance, down from about one third in 2024, yet 54% rate their compliance monitoring capacity as moderate and 42% as developing or limited. Data preparation also remains incomplete: 21% have a data readiness strategy in place and another 41% plan one. Cybersecurity and privacy remain leading concerns, while almost 80% identify IT costs as a risk. Reliance on a small number of external providers raises additional concerns about concentration and dependence. The report identifies organisational controls and data infrastructure as priorities for further readiness. It highlights clear boundaries for AI use and human control, including human approval for policy work, alongside stronger oversight of autonomous AI systems. Diversifying suppliers, adopting interoperable technologies and managing model consumption could help address dependence and costs. On data, priorities include stronger quality controls, metadata and access safeguards so AI systems can securely interpret information. These measures also apply to publicly released central bank data, where inaccurate interpretation by external AI systems could spread misinformation.
Bank for International Settlements survey finds wider central bank AI use but gaps in oversight and data readiness
A Bank for International Settlements survey finds that over two thirds of central banks’ reported AI projects were operational in 2026, versus roughly half in 2024. Wider deployment and stronger governance coexist with limited compliance monitoring, unfinished data preparation and growing concerns about costs and supplier dependence. Priorities include clearer human oversight, supplier diversification and data standards that support secure, accurate use by internal and external AI systems.