BIS Warns of New Risks, Costs and Coordination Needs in AI Adoption by Central Banks

10 October 2025

By David Barwick – FRANKFURT (Econostream) – The Bank for International Settlements on Friday said central banks face major challenges in adopting artificial intelligence, including high costs, cyber vulnerabilities, data governance risks and the need for stronger cooperation to manage technology dependence and talent shortages.

In a report to the G20 on “The use of artificial intelligence for policy purposes,” the BIS said central banks have reported “considerable success” with AI and machine learning, achieving more accurate forecasts and faster analysis. But it warned that the technology’s complexity and opacity make it difficult to interpret model behavior, detect bias or explain decisions to the public.

“Reliable forecasts therefore come at the price of accepting that the underlying model is a ‘black box,’” the report said, noting that this could weaken transparency and communication. The problem is magnified in generative AI, where models risk “hallucinations” and need human supervision for reasoning tasks.

To use AI effectively, central banks must invest heavily in hardware, software, and staff training. Recruiting and retaining specialists who combine economic expertise with coding skills is difficult, the report said, as private demand for such workers remains high.

The BIS said most institutions are likely to rely on “AI copilots” in the near term to assist staff, but should prepare for a second phase in which autonomous “AI agents” handle narrowly defined tasks. Both paths require large-scale retraining, new governance frameworks, and workplace cultures that reward experimentation.

Human oversight will remain essential, it said. “[E]xpert feedback improves models and mitigates hallucinations.” Continuous learning, clear development plans and embedded ethics frameworks can help staff adapt as responsibilities evolve.

Legal and privacy issues are another concern. Central banks must uphold strict data protection standards even as they depend more on private, unstructured data sources such as social media. The BIS cited widespread public concern about privacy breaches and said AI regulation is widely supported but not yet standardized.

AI’s reliance on large datasets also creates governance and interoperability challenges. The BIS urged central banks to strengthen data quality control, auditing and metadata management, adopting FAIR principles—findable, accessible, interoperable and reusable—and collaborating through shared frameworks such as SDMX.

The report warned of concentration risk from dependence on a few large technology providers that control the AI supply chain. Heavy reliance on external models may expose institutions to outages, cyber attacks and reputational damage, while uniform use of top-tier algorithms could lead to herding behavior in stress periods.

Generative AI also heightens cyber threats by enabling more sophisticated phishing, malware and impersonation attacks, the BIS said. Most central banks now see phishing and social engineering as the most likely attack types and have increased cybersecurity investment, staff training and incident-response planning.

Given resource constraints, the BIS urged more cooperation among central banks to share data, models and expertise. Joint procurement, model reuse and open-source repositories can lower costs, reduce environmental impact and accelerate responsible adoption.

In its conclusions, the BIS said the rapid spread of AI means “there is an urgent need for central banks and other supervisory and regulatory authorities to raise their game.” They must strengthen both their role as observers of AI’s economic impact and as users of the technology in their own operations.

The BIS said central banks’ experience with big data and machine learning makes them well placed to lead AI integration, but success will depend on balancing trade-offs between internal and external models, and between collecting in-house data and relying on external providers.

It called for the creation of a “community of practice” among central banks to share knowledge, data and tools. “The BIS is supporting central banks in this endeavor,” the report said.