Learning Objectives

• Identify and describe the risks arising from the widespread use of AI applications in the financial sector.
• Describe how central banks can harness AI to fulfill their policy objectives.
• Explain the macroeconomic impact of AI, including implications for firms’ productive capacity and investment, labor productivity, household consumption, economic output, inflation, and fiscal sustainability.
• Explain how the use of AI presents new opportunities and challenges for central banks, including the central banks’ role as users and providers of data, and the trade-offs posed by their use of both internally-developed and external AI models.

AI Risks in the financial Sector

The widespread use of AI applications in the financial sector brings significant opportunities but also introduces various risks. The key risks arising from AI in the financial sector include:

1. Cybersecurity Risks AI increases exposure to cyber threats, as malicious actors can use AI-powered attacks, such as:
   1. Phishing attacks: AI-generated phishing emails that are more convincing.
   2. Malware creation: AI can generate sophisticated malware to exploit vulnerabilities.
   3. Identity theft and fraud: AI can mimic voices, writing styles, and create deepfake avatars, leading to more advanced scams.
   4. Prompt injection attacks: Manipulating AI models by crafting specific inputs that bypass security measures.
   5. Data poisoning attacks: Malicious actors tampering with the data AI models are trained on, potentially leading to incorrect fraud detection or risk assessments.
   6. Model poisoning attacks: Manipulating AI models during training to introduce vulnerabilities.
2. Financial Stability Risks
   1. Herding behaviour: Since many financial institutions use similar AI-driven models, they may all react in the same way to market events, amplifying volatility.
   2. Procyclicality: AI models trained on past data may reinforce market cycles, leading to more severe booms and busts.
   3. Liquidity crises: Automated trading and AI-driven decision-making can contribute to market crashes if models react similarly in a downturn.
   4. Over-reliance on historical data: AI models may fail to predict unprecedented events, leading to miscalculations in risk management.
   5. Algorithmic collusion: AI models could unintentionally coordinate pricing or market behaviors, leading to anti-competitive practices.
3. Bias and Discrimination AI models trained on biased historical data may reinforce existing inequalities in financial services, such as:
   1. Discriminatory lending decisions: AI could unfairly reject loan applications from certain demographic groups.
   2. Insurance pricing biases: AI could create unfair risk assessments based on variables that indirectly correlate with race, gender, or income level.
   3. Unfair credit scoring: AI may use alternative data (e.g., social media activity) in ways that disadvantage certain populations.
4. Lack of Explainability and Transparency (“Black Box” Problem) Many AI models operate as “black boxes,” meaning their decision-making processes are not easily interpretable by humans. This can create challenges for:
   1. Regulatory compliance: Financial regulators may struggle to assess whether AI-driven decisions are fair and legal.
   2. Customer trust: Consumers may not understand why they were denied a loan or investment opportunity.
   3. Risk management: If financial professionals do not fully understand how AI makes decisions, they may place too much trust in flawed outputs.
5. Third-Party Dependencies and Concentration Risks) The development and deployment of AI in finance are increasingly dominated by a few large technology firms. Risks include:
   1. Single point of failure: If a major AI service provider (e.g., a cloud-based AI model) experiences an outage or a cyberattack, multiple financial institutions could be affected simultaneously.
   2. Regulatory challenges: A handful of firms controlling AI technology could lead to issues related to monopolistic practices and lack of oversight.
6. Data Privacy and Governance Challenges AI relies on vast amounts of data, but financial institutions must adhere to strict data privacy regulations. Risks include:
   1. Data breaches: AI systems storing or processing sensitive financial information could be hacked.
   2. Unauthorized data use: AI models trained on user data could inadvertently violate privacy laws.
   3. Cross-border data governance issues: AI applications operating across jurisdictions may face conflicts in regulatory requirements.
7. Increased Fraud and Financial Crime AI can be used for financial crime prevention (e.g., fraud detection), but criminals can also exploit AI. Examples of AI-enabled fraud include:
   1. Synthetic identity fraud: AI-generated fake identities can be used to open fraudulent bank accounts.
   2. Market manipulation: AI can be used to execute high-frequency trades that manipulate stock prices.
   3. AI-generated misinformation: Deepfake technology could be used to spread false financial information, impacting stock markets.
8. Operational Risks in Financial Institutions AI can be used for financial crime prevention (e.g., fraud detection), but criminals can also exploit AI. Examples of AI-enabled fraud include:
   1. AI system failures: Technical issues, software bugs, or hardware failures could disrupt banking operations.
   2. Over-reliance on AI: If financial institutions rely too heavily on AI without proper human oversight, they may be vulnerable to incorrect predictions or systemic failures.

Harnessing AI By Central Banks For Policy Objectives

Central banks are at the core of economic and financial stability, price stability, monetary policy implementation, payment system oversight, and regulatory supervision. The integration of Artificial Intelligence (AI) and Large Language Models (LLMs) into these functions enables better economic forecasting, financial risk detection, payment system optimization, and cybersecurity enhancement. AI’s ability to process vast amounts of structured and unstructured data in real time has made it a valuable tool for macroeconomic analysis and decision-making. However, successful AI adoption requires strong governance, privacy frameworks, and risk management strategies to mitigate potential threats. The BIS Innovation Hub has launched several AI-driven projects, including Project Aurora, Agorá, Raven, and Neo, to explore AI applications in financial crime prevention, regulatory oversight, payment system modernization, and cybersecurity. Below are the key areas where AI is transforming central banking:

1. Enhancing Economic Analysis and Forecasting
   1. Processing Vast and Diverse Datasets: AI can process and analyze structured and unstructured data far more efficiently than traditional models. This includes:.
      • Macroeconomic indicators: GDP, inflation, employment trends.
      • Financial data: Interest rates, stock prices, and exchange rates.
      • High-frequency data sources: Credit card transactions, online retail activity, and labor market postings.
      • Unstructured data: News articles, corporate earnings reports, social media sentiment, and even satellite imagery.
   2. Real-Time Insights (Nowcasting): AI significantly improves nowcasting by using real-time, high-frequency data to estimate economic conditions more accurately. LLMs trained with financial news and economic reports can extract sentiment indices that help central banks anticipate inflationary pressures and market reactions. For example, an LLM can analyze:
      • Supply chain disruptions from news sources to assess potential price increases.
      • Consumer transaction patterns to detect changes in spending behavior.
      • Wage growth trends from job postings to evaluate labor market strength.
        These AI-enhanced real-time economic assessments allow central banks to respond faster to shifts in economic conditions.
   3. Granular Data Analysis for Targeted Policy Making: AI enables highly detailed tracking of economic conditions across regions, industries, and firms, revealing sector-specific or location-specific trends:
      • AI can track employment trends and wage dynamics using job posting data from online platforms.
      • Electricity consumption and satellite imagery can provide real-time assessments of industrial activity.
      • AI models can analyze household consumption behavior, improving economic forecasting accuracy.
        This granular economic insight allows central banks to design more targeted policies for different sectors and demographics.
   4. Improved Inflation Analysis: Central banks can use AI alongside human expertise to better understand inflationary dynamics. AI models, particularly neural networks, can:
      • Handle significantly larger datasets than traditional econometric models.
      • Identify intricate, non-linear relationships between inflation factors.
      • Capture industry-specific variations in inflation trends. Recent research has decomposed aggregate inflation into four primary sub-components:
        1. Past inflation patterns
        2. Inflation expectations
        3. The output gap
        4. International prices A neural network processes both aggregate (e.g., total services inflation) and disaggregated (e.g., industry-specific output) data to determine each subcomponent’s influence on inflation. This method allows for more precise inflation forecasting and helps central banks respond to inflationary pressures more effectively.
2. Strengthening Financial Stability, Risk Monitoring and Stress Testing
   1. Enhanced Risk Identification and Monitoring: AI enhances financial stability analysis by identifying hidden patterns in large financial datasets. These capabilities allow central banks to:
      
      • Monitor systemic risks across financial and non-financial institutions.
      • Detect anomalies in trading behavior to prevent market manipulation.
      • Assess liquidity conditions and interbank dependencies in real time.
   2. Project Aurora – AI for Financial Crime Prevention: One of the most significant AI-driven projects in central banking is Project Aurora, spearheaded by the BIS Innovation Hub. This project leverages synthetic data in machine learning models, particularly neural networks, to enhance the detection of illicit financial activities such as money laundering, smurfing, and mule accounts. Key insights from Project Aurora include:
      • Neural networks outperform traditional rule-based models in identifying suspicious financial activities.
      • AI models are particularly effective in tracking complex transaction networks, making them valuable tools for anti-money laundering (AML) efforts.
      • Cross-border data pooling across multiple jurisdictions significantly improves detection rates.
      • AI enables real-time identification of suspicious transactions, reducing financial crime risks. This project highlights how AI can support AML (Anti-Money Laundering) and financial stability efforts through automated fraud detection and enhanced transaction monitoring. By automating compliance processes and enhancing transaction monitoring, AI strengthens central banks’ ability to safeguard financial systems from illicit activities.
   3. Macroprudential Regulation and Stress Testing: Pairing AI-based insights with human judgment significantly strengthens macroprudential regulation. Systemic risks often stem from the slow build-up of imbalances and vulnerabilities, materializing as costly stress events. The scarcity of historical crisis data and the uniqueness of financial crises limit the effectiveness of stand-alone, data-intensive AI models. However, when combined with human expertise and informed economic reasoning, generative AI tools can yield substantial benefits for regulators and supervisors. AI-driven stress testing models can simulate real-world shocks, such as interest rate spikes, asset crashes, and liquidity crises—to assess the resilience of the financial system and enhance risk preparedness. In addition, AI-powered early warning systems can identify potential financial crises before they fully materialize by::
      • Detecting hidden risk factors in financial markets.
      • Simulating extreme market shocks using AI-generated synthetic data.
      • Assessing how financial institutions react to various crisis scenarios By integrating AI-driven insights with human oversight, central banks can build robust early warning indicators that alert supervisors to emerging pressure points linked to system-wide risks, thereby reinforcing overall financial stability.
3. Improving Supervision and Regulatory Compliance
   1. Automated Regulatory Reporting and Compliance: AI enhances supervisory oversight by automating regulatory reporting, risk assessments, and compliance monitoring. Key applications include:
      • RegTech solutions that automate compliance processes for financial institutions.
      • AI-driven monitoring of financial transactions to detect market abuse and irregular trading patterns.
      • Machine learning models that analyze corporate balance sheets to assess financial institution health.
   2. Project Ellipse – AI for Real-Time Risk Monitoring: Project Ellipse applies AI to align financial news, economic reports, and corporate disclosures with regulatory risk assessments. This allows regulators to identify emerging threats to financial stability more efficiently.
4. Enhancing Monetary Policy Decision-Making
   1. AI helps central banks assess the effectiveness of monetary policies by:
      • Tracking economic sentiment using AI-powered text analysis.
      • Tracking economic sentiment using AI-powered text analysis.
      • Improving the transmission of monetary policy effects on inflation, wages, and credit markets.
   2. Project Neo – AI for Economic Forecasting: Project Neo applies AI-driven economic forecasting models to build real-time economic indicators for GDP growth, employment trends, and inflation expectations.
5. Strengthening Cybersecurity and Fraud Prevention
   1. As financial systems become more digitized, central banks face increasing cybersecurity threats. AI can play a vital role in detecting and mitigating cyber risks by:
      • Automating routine cybersecurity tasks to improve efficiency.
      • Enhancing threat detection through anomaly detection models.
      • Identifying trends, correlations, and anomalies in cyber threats.
      • Improving training programs to help employees recognize phishing attempt
   2. Project Raven – AI for Cyber Resilience: The BIS Innovation Hub’s Project Raven has demonstrated that AI-powered cybersecurity systems can:
      • Detect and prevent cyber threats faster than traditional methods.
      • Enhance training programs to help individuals identify phishing attempts.
      • Monitor central bank IT networks for potential security breaches.
6. Optimizing Payment Systems and Digital Currencies (CBDCs)
   1. Project Agorá – Revolutionizing Payment Systems with AI: AI is playing a key role in the modernization of payment systems, including the development of central bank digital currencies (CBDCs) and cross-border transaction infrastructures. Project Agorá, another initiative from the BIS Innovation Hub, explores AI-powered tokenization and smart contracts to improve financial transactions by:
      
      • Executing payments atomically to reduce risk.
      • Combining account and messaging updates into a single operation for greater efficiency.
      • Using privacy-preserving resources for AML/KYC compliance to strengthen security.
        This initiative aligns with broader financial modernization efforts, such as Finternet, a collection of interconnected financial ecosystems leveraging unified ledgers and tokenization. These advancements enhance payment security, efficiency, and accessibility while ensuring compliance with regulatory standards.
   2. Fraud Detection and Prevention in Payment Systems: AI enhances fraud detection in financial transactions by identifying anomalies in real-time.
      
      • AI algorithms detect irregular payment behaviors, reducing the risk of identity fraud and cyber theft.
      • AI-driven systems track fraudulent transaction patterns, improving financial system integrity.

Macroeconomic Impact Of AI

The integration of Artificial Intelligence (AI) into the economy has far-reaching implications for firms, labor markets, household consumption, economic output, inflation, and fiscal policy. AI-driven automation, data analytics, and decision-making tools are expected to enhance productivity, reshape labor markets, and influence investment patterns, leading to both opportunities and challenges for central banks and policymakers.

1. Enhanced Productivity and Efficiency:
   
   1. Enhanced Productivity and Efficiency:
      
      • AI adoption increases firms’ productive capacity by automating repetitive tasks, optimizing resource allocation, and improving decision-making. AI-driven technologies, such as machine learning and robotics, enable firms to streamline operations and reduce inefficiencies, leading to higher output per unit of input.
   2. Investment in AI Infrastructure:
      
      • Firms are increasingly investing in AI technologies, particularly in data infrastructure, cloud computing, and automation tools.
      • AI-related investments require substantial upfront costs, but over time, they generate cost savings and efficiency gains.
      • The sectoral distribution of AI investments is uneven, with industries like finance, healthcare, and manufacturing leading in AI adoption.
   3. Potential for Market Concentration:
      
      • Large firms with greater capital reserves may adopt AI faster, potentially increasing market concentration and reducing competition.
      • Smaller firms may struggle to keep pace, leading to productivity disparities across businesses.
2. Impact on Labor Productivity
   
   1. AI as a Productivity Driver: AI contributes to higher labor productivity by:
      
      • Augmenting workers’ capabilities through AI-assisted decision-making.
      • Reducing time spent on routine tasks, allowing workers to focus on higher-value activities.
      • Improving predictive analytics, leading to better supply chain and inventory management.
   2. Skill Shifts and Job Displacement:
      
      • AI adoption is expected to alter the demand for labor, shifting employment patterns from routine tasks to AI-related roles.
      • Demand for high-skilled workers in AI development, data science, and engineering will increase.
      • Low-skilled workers face higher displacement risks, requiring re-skilling and workforce adaptation policies.
   3. Potential for Wage Polarization:
      
      • AI could lead to wage polarization, with high-skilled AI specialists earning premium salaries, while lower-skilled workers face stagnant wages or job displacement.
      • Labor market inequality may widen if AI adoption benefits only specific workforce segments.
3. Impact on Household Consumption 
   
   1. Changing Consumption Patterns:
      
      • AI can increase disposable income by lowering production costs, leading to cheaper goods and services.
      • AI-driven personalized recommendations influence consumer behavior, leading to more targeted spending.
      • AI-powered automation in e-commerce enhances the efficiency of digital transactions, accelerating online shopping trends.
   2. Potential Risks to Employment and Income Stability:
      
      • If AI-driven automation displaces workers, household incomes may decline, negatively impacting consumer spending.
      • Uncertain labor market conditions could lead to higher precautionary savings, reducing aggregate demand.
4. Impact on Economic Output
   
   1. Boosting GDP Growth: AI-driven productivity gains and investment in AI infrastructure have the potential to increase overall economic output by:
      
      • Enhancing firm-level efficiency, leading to higher production capacity.
      • Improving resource allocation across industries.
      • Increasing the speed of innovation and technological progress.
   2. Sectoral Disruptions and Structural Adjustments:
      
      • Certain AI-intensive industries (e.g., tech, finance, healthcare) may experience rapid expansion, while labor-intensive sectors may contract.
      • The economy may undergo structural changes, requiring policy adjustments to support transitioning industries and workers.
5. Impact on Inflation
   
   1. AI’s Dual Effect on Prices:
      
      • Deflationary pressures: AI-driven efficiency gains may reduce production costs, leading to lower consumer prices.
      • Inflationary pressures: If AI increases wage inequality or leads to supply chain disruptions, it could drive sector-specific inflation.
   2. AI’s Role in Inflation Forecasting: Central banks can use AI-powered models to track inflationary trends by:
      
      • Analyzing high-frequency price data to detect inflationary pressures early.
      • Monitoring supply chain disruptions and wage trends.
      • Decomposing inflation into key components (past inflation, inflation expectations, the output gap, and international prices) to enhance forecasting accuracy
   3. AI’s Role in Inflation Forecasting: Central banks can use AI-powered models to track inflationary trends by:
      
      • AI-enabled automation may stabilize inflation by improving production efficiency
      • However, AI-induced labor market disruptions could lead to short-term volatility in wage growth and inflation expectations.
6. Impact on Fiscal Sustainability
   
   1. AI and Tax Revenue:
      
      • AI-driven productivity gains can increase corporate profits, leading to higher tax revenues.
      • However, potential job displacement could reduce income tax collections, requiring fiscal adjustments.
   2. Government Spending on AI Adoption and Workforce Adaptation:
      
      • Policymakers may need to invest in AI education, digital infrastructure, and workforce re-skilling programs to ensure equitable AI adoption.
      • AI can enhance tax collection efficiency through automated fraud detection and improved tax compliance analytics.
   3. Potential Fiscal Challenges:
      
      • Short-term budget pressures may arise from higher social welfare costs due to AI-driven job displacement.
      • Governments may need to adapt tax policies to account for AI-driven economic shifts.

Opportunities and Challenges For Central Banks

The adoption of Artificial Intelligence (AI) in central banking presents both opportunities and challenges, particularly in data management, workforce capabilities, IT infrastructure, and the balance between internal and external AI models. While AI enhances economic forecasting, regulatory oversight, and cybersecurity, its implementation requires significant investment in infrastructure, governance frameworks, and human capital.

New Opportunities for Central Banks in AI Adoption

1. Central Banks as Users, Compilers, and Disseminators of Data –
   
   • Central banks have always played a crucial role in compiling and analyzing economic data. AI enhances this role by improving the ability to process and interpret increasingly large and diverse datasets.
   • AI enables faster, more efficient data dissemination, allowing central banks to provide insights to policymakers, financial institutions, and the general public.
   • The ability to integrate AI with national statistical systems ensures that data remains accurate, comprehensive, and timely.
   • Attackers can create fake video calls with banking clients to bypass traditional identity verification processes.
2. Advancements in Data Governance and Collaboration
   
   • AI facilitates better data-sharing mechanisms between central banks and other institutions, allowing for more efficient financial monitoring.
   • Cross-border cooperation on AI-driven financial regulation can help establish common data governance frameworks that improve global regulatory coordination.
   • AI-powered automation in data curation and metadata standardization can reduce manual processing efforts, ensuring data remains consistent, structured, and accessible.
3. Leveraging AI to Address Rising Commercial Data Costs
   
   • In recent years, the cost of commercial financial data has increased, making AI-driven open data initiatives a cost-effective alternative for monitoring economic activity.
   • AI-powered platforms like BIS Open Tech facilitate international collaboration in financial data sharing and statistical modeling.
   • AI can be used to develop domain-adapted or fine-tuned models specifically for the central banking community, reducing reliance on expensive third-party data providers.
4. Mitigating Environmental Costs Through AI Optimization
   
   • Central banks can optimize AI training processes by reusing trained models and adopting energy-efficient computing methods.
   • Shared AI resources across institutions can reduce redundancy, lowering environmental impact and improving cost-effectiveness.
5. Internally-Developed and External AI Models
   A key decision for central banks is whether to develop AI models in-house or use external AI solutions. Both approaches have advantages:
   • Advantages of External AI Models
     
     • Cost-effective in the short run, eliminating the need for extensive internal development.
     • Leverages private-sector expertise, benefiting from technological advancements and specialized AI research.
     • Faster deployment, as pre-built AI models can be implemented without long development cycles.
   • Advantages of Internal AI Models
     
     • Greater control over data security and privacy, reducing the risk of external data breaches.
     • Higher transparency and explainability, making AI-driven insights more interpretable for regulatory and policy decisions.
     • Customization, allowing models to be tailored to financial stability monitoring, inflation forecasting, and systemic risk assessment.

Challenges of AI Adoption for Central Banks

1. Trade-offs Between Internally-Developed and External AI Models –
   Although there are some advantages with both internally developed and external AI models as mentioned above, they present challenges as well for central banks.
   • Challenges of Using External AI Models:
     
     • Transparency concerns: External models often function as black boxes, making it difficult to audit AI-driven decisions.
     • Dependency risks: Relying on private-sector AI providers creates concerns about vendor lock-in, data access, and pricing power.
     • Security vulnerabilities: External AI models may pose compliance risks if they do not align with central bank security requirements.
   • Challenges of Developing In-House AI Models:
     
     • Significant investment in IT infrastructure is required to support data storage, processing, and model training.
     • Longer development timelines delay AI’s impact compared to off-the-shelf models.
     • Limited scalability makes it harder for central banks to adapt AI models to rapidly changing economic conditions
       ✅Optimal Approach:
       
       • A hybrid AI strategy combining internal model development for critical financial functions with external AI tools for non-sensitive tasks can balance security, efficiency, and innovation.
2. IT Infrastructure and Data Management Challenges – 
   • Scaling Up IT Infrastructure and Energy Concerns:
     
     • AI adoption requires advanced computing power, cloud storage, and high-speed data processing systems.
     • AI models require significant computational resources, leading to high energy consumption.
     • Central banks must invest in secure IT environments to prevent cyber threats and data breaches.
     • Establishing dedicated AI training environments involves high upfront costs, making cost-effective data infrastructure planning essential.
   • Managing Large and Complex Datasets:
     
     • AI models rely on high-quality, unbiased data, but financial datasets are often incomplete, fragmented, or inconsistent.
     • Ensuring data security is crucial when integrating AI with sensitive financial information.
     • AI’s reliance on commercial data vendors has become a challenge, as access to high-quality financial data is becoming increasingly expensive.
       ✅ Solution:
       
       • Developing open-source AI tools for central banks, similar to BIS Open Tech, can provide a cost-effective alternative to commercial data providers.
3. Workforce Challenges: Recruitment, Training, and Retention 
   • Shortage of AI Talent in Central Banks:
     
     • GenAI models operate as black boxes, making it challenging for regulators to trace financial decisions influenced by AI-generated recommendations.
     • The demand for data scientists, AI engineers, and financial technology experts is outpacing supply.
     • Public institutions struggle to compete with private-sector salaries, leading to talent retention issues.
     • Central banks require AI professionals with a strong understanding of economic modeling and regulatory policies, making recruitment even more difficult.
   • Training Existing Staff for AI Integration:
     
     • AI adoption requires retraining central bank employees in data science, machine learning, and financial AI applications.
     • The right mix of technical and economic expertise is needed for AI-driven financial supervision.
       Survey-based evidence suggests that AI-related skill shortages are among the biggest concerns for central banks.
       ✅ Solution:
       
       • Pooling AI expertise across central banks through joint AI training initiatives can help reduce skill gaps and improve workforce readiness.
4. Strengthening Collaboration and Standardizing Data Governance
   
   • Importance of Collaboration in AI Implementation:
     
     • Pooling financial resources across central banks can reduce AI development costs.
     • Sharing AI models and insights can improve efficiency and avoid duplication of efforts.
     • Joint procurement agreements can help lower commercial data costs for smaller central banks.
   • Training Existing Staff for AI Integration:
     
     • Standardized metadata frameworks improve comparability across AI models and reduce data inconsistencies.
     • FAIR (Findable, Accessible, Interoperable, Reusable) principles should be integrated into central bank AI strategies.
     • Privacy-enhancing AI techniques are needed to ensure secure data-sharing between regulatory institutions.
       ✅ Solution:
       
       • A cooperative AI governance approach, including open-source AI models for financial analysis, can enhance cross-border regulatory collaboration.