CIVICA Experts in the Spotlight highlights the research and expertise of academics across our alliance. In this edition, Daria Vernon De Mars and Laura Macchioni from the Florence School of Banking and Finance at the European University Institute explore how artificial intelligence is being used in central banking and financial regulation, and the opportunities and challenges it presents.
Artificial intelligence (AI) is changing how central banks interpret economic developments, how supervisors identify financial risks, and how financial institutions make decisions about access to credit. These developments raise questions that extend beyond technological performance: how can institutions make effective use of AI while ensuring that their decisions remain understandable, accountable and worthy of public trust?
These questions underpin the Report on the CIVICA Workshop on Artificial Intelligence in Finance and Central Banking, published by the European University Institute and bringing together research contributions and practitioner perspectives. The report builds on the workshop hosted at Sciences Po on 29 October 2025, following the Banque de France AI Methods Conference. Developed through a CIVICA thematic activity involving the European University Institute, the Stockholm School of Economics, IE University and Sciences Po, with support from the Young Scholars Initiative, the event connected early-career researchers and practitioners across economics, law, computer science and financial governance.
Across these perspectives, a common insight emerges: realising AI’s potential requires institutions to develop the expertise, organisational arrangements and safeguards needed to use it responsibly.
Expanding the analytical capabilities of central banks
One of AI’s most promising applications concerns the ability to track economic developments more rapidly and precisely. During periods of disruption, conventional indicators may arrive too late to capture changes as they unfold. The report examines how alternative sources, including satellite imagery, electricity consumption and online information, can complement official statistics and support more timely assessments of economic activity.
Research presented in the report illustrates how machine learning (ML) can help interpret satellite observations of air pollution to estimate industrial production, while infrared imagery can provide information on cement production and construction activity. Such approaches are particularly valuable where conventional statistics are delayed or limited. Their usefulness nevertheless depends on data quality and context: alternative indicators can be especially informative during crises, without consistently outperforming established measures in normal conditions.
AI also offers new ways to analyse textual information: for example, large language models (LLMs) can assist researchers in classifying news about monetary policy decisions, expanding databases and checking inconsistencies. These applications illustrate how AI can support analytical work while preserving a role for human judgement in validating and interpreting results.
Strengthening supervision through interpretable tools
Similar opportunities are emerging in supervisory technology, or Suptech. The report describes how machine learning can help identify underreported credit risks, while natural language processing (NLP) supports the analysis of consumer complaints and potential misconduct. Network analysis can also assist authorities in detecting suspicious transaction patterns.
The value of these tools depends on whether supervisors can understand their outputs and incorporate them into their work. Limited technical expertise, legacy infrastructure and data quality problems remain significant constraints. Models trained on historical patterns also require careful interpretation when authorities confront unfamiliar risks.
A contribution on financial networks demonstrates why interpretability matters: by examining secured and unsecured money markets together, researchers identify institutions that connect different market segments and could transmit shocks between them. Their systemic relevance may be difficult to recognise when markets are analysed separately, but using features with clear economic meanings makes these findings more accessible to supervisors and potentially useful for stress testing.
Protecting fairness and accountability in financial decisions
As AI influences credit allocation and customer interactions, its consequences extend to people’s opportunities to obtain housing, finance businesses and withstand economic shocks. The report therefore places inclusion, fairness and the ability to challenge decisions at the centre of financial AI governance.
AI-based credit models can improve assessments of borrowers with limited conventional credit histories. However, historical inequalities and indirect indicators of protected characteristics can also shape their outputs. In fact, greater predictive accuracy does not, by itself, establish that a system treats borrowers fairly or provides meaningful explanations.
Related research on US mortgage markets reinforces the need to examine who benefits from digitalisation. It finds that Fintech lending expands following bank branch closures but does not fully compensate for the withdrawal of bank credit. Although this study concerns digital lending more broadly, it highlights why technological expansion should not be equated with financial inclusion.
These concerns also cross borders: the report’s comparative analysis of AI in credit risk identifies shared commitments to explainability, fairness and human oversight, alongside divergent regulatory approaches. It argues for greater international convergence, while recognising difficult methodological questions, including how to reconcile competing definitions of fairness.
Turning experimentation into institutional practice
Contributions from central banking practitioners show how these issues translate into operational choices, describing how institutions are developing approved AI platforms, controlled experimentation environments and training programmes, supported by legal, cybersecurity and risk expertise.
Applications include summarising documents, analysing market commentary and answering questions using established document collections. More autonomous systems raise further challenges, particularly where tasks require reliable economic reasoning or causal analysis. Moving from a promising pilot to a dependable operational tool consequently requires extensive validation and clear human responsibility.
The report’s cybersecurity contribution develops a related research agenda: examining how oversight, transparency and auditable controls can be embedded in AI-enabled financial security systems. It proposes regulatory sandboxes as settings in which governance arrangements could be tested and refined through structured experimentation.
Taken together, the contributions show why collaboration is central to responsible adoption. Universities provide spaces for critical investigation, while central banks and supervisory authorities bring practical experience of institutional constraints and public responsibilities. Progress will then depend on institutions’ capacity to translate analytical innovation into practices that sustain resilience, fair treatment and public trust.
by Daria Vernon De Mars, Laura Macchioni
About the Authors
Daria Vernon De Mars
Daria Vernon De Mars is a policy professional specialising in cybersecurity and digital governance. She holds an MBA in International Relations and European Studies from the University of Florence and is currently pursuing a PhD in Legal Studies at the same institution. Her research focuses on the regulation of cyberspace, with particular attention to experimentalist governance approaches.
She is currently part of the Single Supervisory Mechanism (SSM) team at the Florence School of Banking and Finance, where she develops executive training offerings on digital finance.
Before moving into academia, she gained significant experience in the private consulting sector as a Cyber Security Consultant for financial industries and government entities.
Laura Macchioni
Laura Macchioni is a Research Associate at the Florence School of Banking and Finance (FBF), where she contributes to the teaching and training activities of the EU Supervisory Digital Finance Academy. She also leads a European Commission Technical Support Instrument (TSI) project on financial literacy and artificial intelligence. Her research focuses on digital finance, with particular interests in tokenisation, anti-money laundering and AI.
Laura holds a degree in International Economic and Business Studies (Mag.rer.soc.oec.) from the University of Innsbruck and a postgraduate Master's degree in European Studies (M.E.S.) from the University of Vienna.
Before joining FBF, she worked in the European Banking Authority's Economic and Risk Analysis Department in Paris, where she contributed to the organisation of the EBA's annual Policy Research Workshop and supported task force activities related to the Basel framework.
