AGP Picks
View all

Financial stability implications of AI adoption in the financial industry and the real economy

News originally published by

ECB - European Central Bank

Prepared by Leonardo Andaloro, Maria Leonor Puga, Sándor Gardó, Benjamin Klaus, Urban Rupnik and Christian Weistroffer[1]

Published as part of the Financial Stability Review, November 2026.

The rapid adoption of artificial intelligence (AI) is reshaping the financial industry as well as the broader economy, creating both opportunities and risks. Banks and asset managers, which represent key channels of financial intermediation in the euro area, are increasingly using AI for tasks such as credit scoring, fraud detection and portfolio management. While AI adoption can support innovation and enhance operational efficiency, it may also introduce new risks and amplify existing vulnerabilities, including a reliance on a small number of technology providers that creates concentration and third-party dependencies across the financial system. At the same time, financial institutions are also increasingly exposed to AI-driven developments in the real economy. A novel firm-level indicator suggests that a sizeable share of firms, concentrated mainly in the IT and financial sectors, is significantly exposed to AI-related opportunities and risks. Policies should focus on improving the quality of data that monitor the use of AI, strengthening the oversight of critical AI providers and adapting market safeguards and supervisory capabilities to emerging risks.

1 Introduction

The rapid adoption of artificial intelligence is transforming the financial sector and the economy in general, creating both opportunities and risks that are relevant for financial stability. Since the launch of ChatGPT in November 2022, advances in generative artificial intelligence (GenAI) and, increasingly, agentic AI have expanded the range of applications and lowered the barriers to adoption. This has the potential to boost productivity, reshape business models and intensify competition. Firms, including financial institutions, are investing heavily in AI infrastructure and applications. AI could improve information processing, risk assessment and operational efficiency, enabling better use of large, complex datasets and faster responses to evolving conditions. However, the speed of adoption, the growing autonomy of AI systems and the reliance on a small number of technology providers is also raising new questions for financial stability.

The financial stability implications of AI reflect both its use in specific sectors and broader risks that cut across the financial system. From customer service and fraud detection to underwriting, research and trading, AI is permeating nearly all aspects of financial activity. However, the implications of AI adoption for financial stability vary significantly across different types of financial institution, reflecting each institution’s distinct business model and operational priorities. This is why a sector-specific evaluation is particularly relevant. Banks and asset managers are emerging as natural focal points for this analysis, as they represent the two primary channels of financial intermediation in the euro area: dominant bank-based finance and growing market-based finance. Beyond the financial sector itself, the adoption of AI by non-financial corporations is introducing additional risks that affect all financial institutions to a similar extent. Such risks include relying on a concentrated group of AI model and cloud service providers as well as exposures to firms undergoing AI-driven business transformations. These interconnected risks underscore the need to assess AI’s impact on financial stability, both at the sectoral level and across the financial system as a whole.

This special feature assesses the implications of AI adoption for financial stability, with a focus on banks and asset managers. Building on previous work done by the ECB and drawing on recent survey data, bank and fund-level evidence and emerging research, the special feature provides new insights into the rapid adoption of AI and the implications for financial stability.[2] The analysis first explores AI’s adoption and use across countries and economic sectors. It then turns to how financial institutions – particularly banks and asset managers – are integrating AI into their activities and the associated opportunities and risks. Finally, it examines risks arising from concentration and dependence on AI providers, as well as those stemming from financial sector exposures to firms undergoing AI-driven transformation. It concludes with policy considerations on mitigating the risks and meeting the challenges posed by AI adoption.

2 AI adoption trends across the globe and the broader economy

AI adoption has surged globally in recent years, although progress within the EU has been more modest and uneven. AI has moved beyond the experimental phase to become a widely used organisational tool within a remarkably short period of time. According to Eurostat, however, the rate of AI adoption by firms remains modest within the EU, with significant differences across countries and firm sizes. By 2025 around 20% of EU enterprises were using at least one AI technology, up by some 6.5 percentage points from 2024. Adoption rates vary widely, with northern countries leading the way and southern and eastern European countries lagging behind (Chart A.1, panel a).

Sectoral adoption is highly concentrated in data-driven activities, with increasing automation potentially amplifying systemic risks. More granular US data show that IT and communications, finance and professional services report the highest levels of AI adoption, while sectors such as manufacturing report significantly lower levels (Chart A.1, panel b). Even in finance, however, uptake varies considerably (e.g. between securities trading and banking). In general, AI is embedded in data-driven functions and is concentrated in large firms and high-value services rather than spread evenly across the economy. Moreover, AI adoption is not just broadening, it is also deepening. The Anthropic Economic Index indicates that, between January and August 2025, automation in task-oriented interactions on Claude.ai, such as coding or financial modelling, rose from 27% to 39% of all interactions.[3] This signals a shift from assistive to semi-autonomous use that could reshape front-office and risk functions in finance.

Chart A.1

AI adoption has increased rapidly over the last few years, but progress remains uneven across EU countries and across US sectors of economic activity

a) AI adoption across EU firms, by country

b) AI adoption across US economic sectors

(2024-25, percentages)

(Feb. 2026, percentages)

Sources: Eurostat, US Census Bureau and ECB calculations.
Notes: The panels show AI use/adoption reported in publicly available surveys. Definitions, survey populations and question wording differ across sources and, in some cases, across survey waves, meaning that levels are not directly comparable. Panel a: the chart depicts the share of enterprises in selected EU countries with ten or more employees and self-employed persons using at least one AI technology, as defined by Eurostat. These technologies include text mining, speech recognition, natural language generation or speech synthesis, image/video/audio generation, image recognition, machine learning, AI-based workflow automation or decision support, and autonomous machines. The euro area aggregate represents the share across all eligible enterprises, weighted by the number of enterprises. Panel b: the chart shows the share of US businesses in selected sub-sectors reporting AI use in any business function. Owing to differences in coverage and wording, levels are not comparable across both panels. “Finance & ins.” stands for finance and insurance; “Info. & comm.” stands for information and communications; “Prof. & tech.” stands for professional and technical; “Real est. & leasing” stands for real estate and leasing; “Arts & rec.” stands for arts and recreation; “Real est. and leasing” stands for real estate and leasing; “Computers and elec.” stands for computers and electronics.

3 AI adoption and use by banks and asset managers

AI has significant potential to improve the efficiency and quality of financial intermediation, but also to alter and magnify existing risks. Current AI use cases differ markedly across business models (Figure A.1). In banks, AI enables faster processing of large datasets, more granular borrower assessment and enhanced detection of suspicious activities. In markets and asset management, AI applications can reduce information-processing costs, accelerate research and enhance portfolio optimisation and customer relationships. Such benefits help explain why AI adoption is rising across a financial industry in which data-intensive and judgement-based tasks are prevalent. Despite the differences in use cases, the underlying technology and third-party dependencies expose financial institutions to common risks, as their processes rely on AI-powered technology. Risks include data privacy issues, model reliability as well as alignment, explainability and accountability concerns, cyber and malicious-use risks and risks from reliance on third-party providers.

Beyond these technological risks, the rapid proliferation of AI applications is creating challenges for managing AI risk on a larger scale. As firms move on from experimentation with AI to the industrialisation of AI across business processes, they need consistent frameworks and guardrails governing its use. These challenges will become more pronounced as AI systems gain greater autonomy and move from supporting analysis towards preparing or executing decisions, including through agentic AI. Cost considerations are also increasingly important, as firms need to weigh the greater capabilities of more expensive frontier models against cheaper alternatives, as well as the relative cost and value of AI and human labour.

3.1 AI adoption in the banking sector

Euro area banks have accelerated their adoption of AI, moving swiftly from experimenting with it to deploying it operationally. This trend is particularly pronounced for generative AI (GenAI), while agentic AI remains largely in the pilot or planning stage for most global financial institutions surveyed (Chart A.2, panel a). Zooming in on euro area banks, the share of such banks actively using GenAI almost doubled between 2024 and 2025, increasing from 46% to 86%, while the proportion still in the development phase dropped from 38% to 10% (Chart A.2, panel b). This shift not only reflects the broader deployment of GenAI but also its growing relevance. In fact, the share of banks assigning high business relevance to their GenAI applications has also grown, highlighting their increasing relevance for core business functions. IT operations and legal services are the leading generative AI use cases and have been adopted by roughly 50% of euro area banks. If applications under development are included, this figure rises to 80%. Adoption is less advanced in areas with more complex requirements, such as risk management (Chart A.2, panel c).

Chart A.2

Euro area banks have markedly increased their use of generative AI in recent years, primarily for IT operations, legal and regulatory activities as well as frontline services

a) Degree of development of AI systems across global financial institutions

b) Use of GenAI across euro area banks

c) GenAI use cases across euro area banks

(Oct. 2025, percentages)

(2024, 2025, percentages)

(2025, percentages)

Sources: Renier et al.*, ECB (supervisory data) and ECB calculations.
Notes: Panel a: the data shown in this chart are based on a survey conducted by the Institute of International Finance and Ernst & Young Global Services. There were 61 participating financial institutions (banks, insurers, asset managers and capital market firms) headquartered in eight global regions. GenAI is defined in the survey as “Systems that create new content – including text, audio and video – based on their training data and in response to prompts”, while agentic AI is defined as “AI systems that use sophisticated reasoning and iterative planning to autonomously solve complex, multi-step problems. These agents have a diverse range of applications and perform autonomous tasks on behalf of the user or another system”. Panel b: the chart shows the use status of GenAI as reported by euro area banks, where the status “in use” is broken down by business relevance; the status “development” includes exploration, development and testing. Panel c: the chart shows the use status of the different prevailing GenAI use cases as reported by euro area banks, where the status “development” includes exploration, development and testing.
*) Renier, J., French, C., Bulchandani, L. and Mendez Delgado, D., “IIF-EY Annual Survey Report on AI Use in Financial Services”, Institute of International Finance, 16 October 2025.

Euro area banks identify data challenges, model reliability, explainability and third-party concentration as key risks associated with AI adoption. When adopting GenAI, financial institutions cite data confidentiality and model hallucinations as their top concerns. Agentic AI is seen as introducing distinct risks, such as autonomy, model drift and lack of explainability of actions (Chart A.3, panel a). In addition, although several banks reported that they relied on multiple vendors and on alternative providers and internal tools, AI adoption by euro area banks depends heavily on a small number of US-based technology providers (Chart A.3, panel b). A downside of this reliance on a small number of hyperscalers heightens exposure to correlated errors and vendor-specific shocks.[4] Furthermore, cyber and malicious-use risks, such as prompt injection,[5] deepfakes and AI-enabled phishing, could increase operational risk. The potential systemic dimension of these risks has been highlighted by the ESRB’s warning on systemic cyber risks stemming from frontier AI models[6], and supervisors have reinforced their expectations on cyber and operational resilience accordingly. Looking ahead, financial institutions are expecting regulatory and supervisory initiatives to increasingly shape how they develop and deploy AI systems. Regulators are perceived as having greater experience in more established fields of expertise such as AI governance and data frameworks. By contrast, surveyed institutions perceive regulatory understanding of agentic AI or open-weight models to be much weaker (Chart A.3, panel c), highlighting the challenges posed by the rapid pace of technological innovation.

Chart A.3

GenAI adoption risks include data issues, model reliability and reliance on US providers, while institutions see distinct risks and oversight gaps in agentic AI

a) Top concerns in GenAI and agentic AI applications

b) GenAI vendors and service providers used by euro area banks

c) Share of firms saying their regulator understands the area of the AI ecosystem

(Oct. 2025, percentages)

(2025, percentages)

(Oct. 2025, percentages)

Sources: Renier et al.*, ECB (supervisory data, SREP) and ECB calculations.
Notes: Panel a: each of the participating firms were asked what they considered to be the top three risks/concerns in using GenAI and agentic AI applications. Panel b: the chart is based on banks’ free-text comments regarding the external developers or vendors for generative AI sourcing as well as third-party service providers for generative AI service dependence. Mentions of Microsoft and OpenAI products and services, including Azure, OpenAI, ChatGPT, GitHub Copilot and Microsoft 365, are grouped under “Microsoft/OpenAI”. This grouping reflects the partnership between the two companies and the integration of OpenAI models into several Microsoft offerings. References to Google-related services, including Google Cloud and Gemini, are grouped under “Google”, while references to Amazon Web Services, including Bedrock, are grouped under “AWS”. Mentions of banks’ own, proprietary or private-cloud solutions are classified as “Internal tools”. References to other AI model providers, such as Mistral, Claude or Anthropic, and to other software or platform providers are combined under “Other tools”. Responses indicating that no provider is used, that the question is not applicable or that no specific provider can be identified are classified as “Not specified”. Comments may be assigned to more than one category if they refer to multiple providers or tools. Panel c: firms were asked in which of the listed areas they believe the regulator/supervisor is equipped to understand the AI ecosystem.
*) Renier, J., French, C., Bulchandani, L. and Mendez Delgado, D., “IIF-EY Annual Survey Report on AI Use in Financial Services”, Institute of International Finance, 16 October 2025.

3.2 AI adoption in the asset management industry

AI adoption is accelerating in asset management, with GenAI used mainly in research and operational tasks rather than in direct investment decisions. Recent evidence indicates that asset managers in the EU are increasingly adopting AI, with these technologies gradually being integrated into a broad range of securities-market activities.[7] These technologies range from established machine-learning techniques to more recent applications of generative and agentic AI, which differ in their capabilities and maturity. Generative AI is increasingly being used to process information, analyse company reports, support investment research, generate code and optimise internal workflows. While its adoption is becoming widespread, its use remains largely focused on research and operational tasks, supporting rather than replacing human judgement.[8] By contrast, established machine-learning techniques are being used more directly in investment processes, such as analysing large volumes of market and trading data, identifying patterns and investment signals, and forecasting volatility and liquidity. AI-generated insights are informing key activities such as security selection, portfolio construction, trading strategies and robo-advice. However, more sophisticated applications remain concentrated among a relatively limited number of asset managers with the necessary data infrastructure, technology and quantitative expertise.

While AI adoption could bring significant benefits, it could affect financial stability by reshaping market structures. On the one hand, AI adoption could, in principle, reduce operational costs and enhance market efficiency through better information processing, risk assessment and diversification (Figure A.2). It could also lower barriers to entry and support the expansion of smaller and more specialised firms. On the other hand, economies of scale in AI adoption could favour larger asset managers that are better positioned to make substantial upfront investments in the necessary data infrastructure, skilled personnel and effective oversight, potentially increasing market concentration. At the same time, as in the banking sector, relying on a small number of providers for hardware, cloud computing and pre-trained models is creating common operational dependencies.[9] This is raising concerns related to third-party concentration in securities markets that could amplify the impact of operational disruptions.[10]

Figure A.2

AI in asset management could enhance market efficiency, but may also increase concentration risk and amplify propagation of shocks

Source: ECB.

The growing use of similar AI models and investment signals could increase correlated trading and amplify the propagation of market shocks. A number of investment funds use machine-learning techniques to identify trading opportunities. These AI-driven funds can be identified in Morningstar data using an indicator that captures the use of machine learning in investment strategies. AI-driven funds tend to attract stronger inflows, trade more frequently and generate slightly higher returns net of fund-level costs than their non-AI peers (Chart A.4, panels a and b), while still accounting for a relatively small share of the market.[11] Evidence is also emerging of more correlated trading among AI-driven equity funds, although their responses to market stress may vary.[12] Moreover, AI-driven equity funds are showing a persistently higher downside tail dependence than comparable non-AI funds, suggesting a greater tendency to experience large losses simultaneously (Chart A.4, panel c). While these findings reflect predominantly established machine-learning applications, they may provide insights into risks that could become more pronounced as advanced AI technologies are increasingly integrated into investment decisions. At current levels of adoption, these effects are unlikely to pose systemic risks. However, they could become more significant if AI-driven strategies gain substantial market share, especially in specific market segments or among leveraged investors. Under such conditions, correlated behaviour could accelerate shock propagation and amplify market stress, including through synchronised deleveraging and fire sales.[13]

Chart A.4

AI-driven funds attract stronger investor flows and trade more frequently, while showing greater downside tail dependence

a) Cumulative monthly fund flows relative to initial fund size

b) Turnover ratio and annualised return across asset classes

c) Likelihood of co-movement during negative return episodes

(Jan. 2023-Sep. 2026, percentages)

(Jan. 2023-Dec. 2025, percentages)

(1 Nov. 2025-26 Oct. 2026, percentages)

Sources: Morningstar DBRS and ECB calculations.
Notes: The analysis draws on a global sample of investment funds across different asset classes, excluding passive strategies, comparing machine-learning and non-machine-learning funds using propensity score matching. Panel a: the chart shows average cumulative investor flows for AI and matched non-AI funds. Monthly net flows are calculated as a percentage of fund assets at the preceding month, winsorised at the 1st and 99th percentiles, compounded over time and then equally averaged across funds. Panel b: the chart illustrates the average three-year turnover ratio (x-axis) and annualised three-year return (y-axis) of funds in the sample for AI and non-AI funds, per asset class. Panel c: the chart depicts the mean probability that an equity fund is in its own lower return tail, conditional on a peer fund already being in its lower tail, calculated using daily returns from 1 November 2025 to 26 October 2026 and averaged across within-group fund pairs. The x-axis represents different percentile thresholds in the lower tail of each fund’s return distribution, while the y-axis shows the corresponding mean empirical lower-tail dependence coefficient. The difference in lower-tail dependence between AI and non-AI funds is positive across all plotted thresholds and statistically significant under a joint directional test covering the 10%, 15% and 20% thresholds (p = 0.049).

Faster information processing and automated responses to market-relevant news could accelerate and amplify short-term price moves. While current generative AI models are generally too slow for high-frequency trading, they could still substantially shorten the time needed to analyse information and translate it into investment decisions. AI models could, in principle, respond not only to changes in fundamentals but also to rumours and shifts in market sentiment, potentially reinforcing short-term price adjustments and leaving less time for market participants and authorities to react. Such risks may become more relevant as brokerage platforms introduce agentic trading capabilities. These tools, which are still at an early stage, allow AI-enabled trading agents to monitor markets and portfolios and execute trades on behalf of users, including in equities, options and crypto markets.[14] Their take-up and market impact are not yet known. Looking ahead, the use of reinforcement learning, in which trading agents learn strategies autonomously by optimising profit, raises additional concerns. Research suggests that such agents may learn manipulative behaviours, such as spoofing, predatory trading or tacit collusion, without being explicitly programmed to do so. As this evidence stems mainly from simulated settings and the use of such agents in live trading appears to be limited, this risk is currently more relevant for market integrity than for financial stability.[15]

4 AI risks to the financial system from outside the sector

4.1 Concentration in the AI supply chain

Beyond the medium-term risks of AI integration, financial institutions face more immediate operational risks arising from third-party dependence. Supply chain concentration poses risks, as a small number of dominant cloud and model providers could exert outsized influence. This could raise switching costs, increasing the chances of outages, model updates, pricing changes or application programming interface (API) restrictions affecting many financial institutions simultaneously. As mentioned above, similar risks arise in financial markets, where widespread use of common models and data can increase herding and correlated trading. Growing third-party dependence could also amplify cyber risks, as vulnerabilities in external vendors’ systems could become entry points for attacks. At the same time, large cloud providers may offer greater protection against cyberattacks, given their substantial resources and expertise.

Concentration in the AI supply chain could allow operational incidents at a small number of providers to escalate into system-wide disruptions. The broad European market for two pillars of the AI stack, cloud computing and GenAI, is dominated by US firms. Amazon Web Services (AWS), Microsoft and Google accounted for 82% of the cloud market and, together with OpenAI, 90% of GenAI services in 2025 (Chart A.5, panel a). While this structure supports economies of scale, rapid innovation and access to advanced infrastructure, it also increases dependency on a narrow set of vendors, cloud regions, managed services and model providers.[16] As financial institutions embed AI into customer service, fraud detection, compliance, software development and risk management workflows, they rely increasingly on the same upstream providers. Outages, cyber incidents, model failures or contractual disruptions could thus become more correlated across firms.

Chart A.5

Heavy reliance on a few US cloud and GenAI providers heightens the risk of a single outage having widespread effects across sectors and national borders

a) Cloud computing and GenAI market shares in Europe

b) User-submitted outage reports related to disruptions at AWS on 20 October 2025, by service type and country

(2025, percentages)

(Oct. 2025, percentages)

Sources: Gineikyte-Kanclere et al.*, Downdetector by Accenture** and ECB calculations.
Notes: Panel a: estimated cloud market shares in Europe based on publicly available market intelligence and company reporting, with approximate shares for leading cloud service providers. The estimates cover the broad European cloud market, including infrastructure services and, where relevant, major software-as-a-service activities. Estimated EU market shares for selected GenAI providers based on available indicators of chatbot usage, subscription and API revenues, enterprise adoption and foundation model platform activity. Panel b: the chart shows disruptions in consumer apps, web portals and AWS’s own services, grouped by affected service type and country based on global user reports.
*) Gineikyte-Kanclere, V., Eggert, M. and Skiotyte, G., “European Software and Cyber Dependencies”, Policy Department for Transformation, Innovation and Health, European Parliament, December 2025.
**) See Kehoe, L., “Revealing the Cascading Impacts of the AWS Outage”, Ookla, 22 October 2025.

The October 2025 AWS outage illustrates how the effects of a localised technical failure can propagate rapidly across sectors and borders. The disruption, centred on the US-East cloud region, affected services ranging from social messaging and gaming to e-commerce and banking. The United States accounted for the largest share of user-submitted reports, but disruptions were recorded in more than 60 countries, and especially in the United Kingdom, Germany and the Netherlands. This underscored the cross-border nature of cloud dependencies (Chart A.5, panel b). Although financial institutions accounted for only a small share of reports, the incident illustrates the potential financial stability implications of such a disruption, as the financial industry relies on the same digital infrastructure as the wider economy. The primary concern is not that a single outage could directly threaten the solvency of financial institutions. It is more that concentrated AI and cloud dependencies could weaken operational resilience, amplify confidence effects and complicate recovery when multiple critical services share a common point of failure. This highlights the need for dependency mapping, multi-region and multi-provider resilience, exit strategies and robust oversight of cloud and AI providers.

4.2 Spillovers from AI-driven changes in the real economy

Beyond the direct use of AI in the financial sector, financial stability could also be indirectly affected by AI-driven changes in the real economy. A novel firm-level indicator of AI intensity suggests that these effects could be material, although they would be unevenly distributed across the non-financial corporate sector. Around 20% of firms enjoy mainly gains from AI, 19% experience both gains and threats, 60% have a low level of gains and threats, while hardly any firms face mainly threats (Chart A.6, panel a). This implies that a sizeable share of lending and securities exposures could be linked to firms whose revenues, costs, margins or competitive positions could change as AI adoption spreads. For firms with mainly gains, AI may support productivity, profitability and debt servicing capacity.[17] For firms experiencing higher levels of gains and threats, the outcome is likely to depend on how they adapt. Some firms that adopt AI could strengthen their positions, while others could see demand for their products and services, their prices or their role as intermediaries come under pressure from AI-enabled competitors, weakening their business models and increasing their risk of default.

Chart A.6

Sizeable parts of the global corporate sector are likely to be transformed by AI, particularly in information technology and finance

a) Decomposition of AI intensity categories of firms

b) AI intensity of firms across sectors of economic activity

(share of firms)

(index)

Sources: Moody’s, ECB (RIAD) and ECB calculations.
Notes: The AI intensity indicator is based on firm-level assessments of each company’s business description in Orbis, generated by a language model (a FinBERT model trained on assessments by a large language model for a stratified sample of 100,000 companies). The sample covers around 4.5 million firms (44% in Asia, including 28% in China; 33% in Europe, including 18% in the euro area; and 16% in North America). Firms are scored along two dimensions: AI benefit (the extent to which AI can improve the firm’s work and products) and AI risk (the extent to which AI threatens the demand for, the price of or the role of the firm’s products or services). For each dimension, the model assigns a level of low, medium or high, mapped to values of 0, 0.5 and 1. Public administration, other services, residential care and social work, households as employers and extraterritorial organisations are excluded, as are holding companies, shell companies and funds. Firms and sectors are divided into four groups by lines drawn at the average AI benefit and AI risk of all firms shown: mainly gains (above-average benefit, below-average risk), gains and threats (both above average), mainly threats (below-average benefit, above-average risk) and low gains and threats (both below average). Panel a: share of firms in each group, assigned at firm level. “Mainly threats” accounts for less than 0.5% of firms and is not shown. Figures may not add up to 100% due to rounding. Panel b: bubbles show sector averages of AI benefit and AI risk. Bubble size reflects the number of firms. Sectors are NACE Rev. 2 sections assigned from the activity described by each firm rather than its registered code. “Health & pharma” combines human health activities with the manufacture of pharmaceuticals and medical devices.

The sectoral distribution of AI-driven opportunities and risks will likely determine the extent to which corporate disruption affects financial institutions. Most sectors of economic activity remain in the low gains and threats quadrant, suggesting limited immediate exposure to current AI capabilities (Chart A.6, panel b). By contrast, IT and financial firms stand out at the other end of the spectrum, with significant potential gains but also above-average threats. For the financial sector, risks may arise not only from direct exposures to disrupted firms, but also from longer-term transformative effects. AI-driven productivity shocks could alter relative prices, wages and employment across sectors and regions, while structural labour market changes could affect household income, consumption and mortgage performance. At the same time, productivity and profitability gains concentrated among large, digitally advanced firms could intensify competitive pressure on smaller or less adaptable firms. These dynamics may well unfold gradually, but they could still reprice credit and equity risk if investors reassessed which firms are likely to benefit from, or be displaced by, AI.

Euro area financial institutions’ exposures to AI-exposed firms vary considerably across instruments and sectors. Non-bank exposures to AI-intensive firms are significant and growing, particularly through investment funds’ equity holdings and, to a lesser extent, debt securities, while bank lending remains limited (Chart A.7, panels a and b). This suggests that any reassessment of the earnings prospects of firms whose business models are affected by AI would initially be transmitted mainly through market price corrections and valuation losses for non-bank investors, rather than through bank credit losses. At the core of exposures to AI-intensive sectors are big tech and specialised AI firms that are investing heavily across the AI value chain, including in data centre infrastructure and AI model development and training. More recently, the rapid increase in AI-related spending has led to greater reliance on debt securities and other forms of external financing, including off-balance-sheet financing backed by future revenues. Financial sector debt exposures to firms in the AI supply chain have increased as a consequence, with a substantial share concentrated in a few hyperscalers (Chart A.7, panel c). This could broaden the channels through which a reassessment of AI-related earnings prospects could affect the financial system, extending potential losses beyond equity investors to credit providers. Growing leverage in non-banks and concentrated exposures to AI-intensive sectors could further amplify vulnerabilities.

Chart A.7

Non-bank exposures to AI-intense firms have increased, especially in equity markets, while bank exposures remain more limited

a) Financial sector exposure
to AI via lending and securities

b) Holdings of debt and equity securities issued by AI-intensive firms

c) Exposure to firms in the AI supply chain

(Q2 2026, € trillions)

(Q1 2020-Q2 2026, € trillions)

(Q1 2025, Q2 2026; percentages)

Sources: Moody’s, ECB (RIAD, AnaCredit, SHS, supervisory data) and ECB calculations.
Notes: Panel a: the chart shows euro area sector exposures through lending and securities holdings to firms classified by the AI intensity index. Exposures to borrowers and issuers not linked to a scored firm are allocated across the four groups based on the composition of scored firms within their respective NACE Rev. 2 section. Exposures to holding companies, head offices and counterparties without a NACE code are reported as “not classified”. Lending data cover outstanding loans from euro area credit institutions to non-financial corporations, with borrowers linked to scored firms through RIAD-Orbis mapping. Securities data cover euro area sector holdings of debt and equity securities issued by non-government entities. ICPFs stands for insurance corporations and pension funds; IFs stands for investment funds; OFIs stands for other financial intermediaries. Panel b: the chart includes firms classified in the “Gains and threats”, “Mainly threats” and “Mainly gains” categories according to the AI-intensity index. Panel c: firms in the AI supply chain comprise 168 firms and their subsidiaries involved in AI model development, AI chip production, cloud provision, data centre and related infrastructure, or utilities. The hyperscaler subset comprises Alphabet, Amazon, Broadcom, Meta, Microsoft, Nvidia and Oracle. The y-axis shows debt of firms in the AI supply chain as a share of each financial sub-sector’s total debt holdings, while the x-axis shows the share of this exposure concentrated in hyperscalers. The bubble size represents the absolute amount of AI supply chain debt held by each sector, in billions of euro. Solid bubbles denote Q2 2026 and light bubbles denote Q1 2025, with lines tracing each sector’s movement over the year.

5 Conclusions

The financial stability implications of AI are likely to grow as its use becomes more widespread and increasingly autonomous across the financial sector and broader economy. GenAI adoption is spreading rapidly across the financial sector, while more advanced applications in specific areas such as credit assessment, portfolio management and trading remain less widespread, with agentic AI still largely at the pilot stage. As AI becomes more deeply embedded in financial activities, greater automation and reliance on a concentrated set of technology providers could increase operational risks, while AI’s more widespread use in investment and trading could contribute to correlated behaviour in financial markets. At the same time, AI is creating both opportunities and risks for firms across the broader economy, which could affect financial institutions through their credit and equity exposures. If properly managed, AI adoption could also support the EU’s broader digitalisation and productivity objectives. By reducing information and transaction costs, improving investment analysis and enhancing the efficiency of financial intermediation, AI could contribute to more efficient and accessible capital markets, complementing the objectives of the savings and investments union.

Managing emerging AI-related risks requires robust governance, close monitoring and international cooperation. As firms move from experimentation towards the broader deployment of AI across business processes, consistent governance frameworks and guardrails will play an increasingly important role in managing the associated risks, as also emphasised in the Financial Stability Board’s sound practices for the responsible adoption of AI.[18] Greater transparency from AI model providers could help financial institutions assess model risks and third-party dependencies. At the same time, authorities and central banks will need to closely monitor the evolving implications of AI adoption for financial stability. Such monitoring should take a system-wide perspective and be supported by close international cooperation, given the global nature of AI adoption and the potential for cross-border spillovers.

Several EU policy and supervisory initiatives already address important aspects of AI-related risks, but continued efforts are needed to keep pace with technological developments. Model risks, particularly those related to explainability and reliability, are being addressed through prudential supervisory expectations for AI-based internal models.[19] Third-party dependencies and concentration risks are covered by DORA[20] and complemented for some AI-specific risks by the AI Act.[21] The AI Act also provides cross-sectoral safeguards, including for certain applications in financial services, such as creditworthiness assessments. Moreover, AI-enabled cyber risks and their potential systemic implications have been explicitly highlighted by the ESRB’s warning on systemic cyber risk stemming from frontier AI models, with supervisors reinforcing their expectations regarding cyber and operational resilience.[22] To effectively address these challenges and safeguard financial stability, regulators, supervisors and other relevant authorities need to collaborate to ensure the effective implementation of existing and newly established frameworks, and continue to evaluate emerging gaps, while strengthening their own technological expertise to keep pace with the evolving use of AI in the financial sector.

Legal Disclaimer:

EIN Presswire provides this news content "as is" without warranty of any kind. We do not accept any responsibility or liability for the accuracy, content, images, videos, licenses, completeness, legality, or reliability of the information contained in this article. If you have any complaints or copyright issues related to this article, kindly contact the author above.

Share this page:

Advanced Search Options

Search for:

Search scope:

Type:

Search in:

Date range:

The last

Sort by:

Sign up for:

Fintech World Post

The daily local news briefing you can trust. Every day. Subscribe now.

By signing up, you agree to our Terms & Conditions.