Artificial intelligence

Latest authority-led adoption studies point to a continued upward trend in AI uptake across financial services, with several jurisdictions now reporting adoption levels of up to 90% or more among surveyed firms. However, adoption remains uneven across sectors, institutions and use cases: banks, insurers and larger firms generally lead, while asset managers, smaller firms and some non-bank segments show more mixed uptake and earlier-stage maturity. Generative AI is acting as a key accelerator, particularly for internal productivity, knowledge work and data processing applications such as drafting, summarisation, translation, coding support, internal search, research assistance and call centre or adviser support. Customer-facing, investment, trading and core financial decisioning uses are emerging, but remain more bounded, specialised and subject to human validation. At the same time, agentic AI is beginning to appear as a more autonomous layer for multi-step workflows, although adoption is still early and generally constrained by controls. Across the sector, the main challenge is no longer whether firms are experimenting with AI, but whether they can scale it safely and demonstrate business value amid persistent barriers around ROI, data readiness, third-party dependencies, cross-border governance fragmentation and shortages of hybrid AI, regulatory and risk-management talent.

Latest evidence based on authority-led adoption studies published during 2026 points to a continued upward trend in AI uptake, several jurisdictions reporting uptake of up to 90%+ across financial services firms. Adoption however continues to remain uneven: banks, insurers and larger institutions generally lead while asset managers and smaller or non-bank segments show more mixed uptake and more early-stage maturity.

Generative AI remains a key accelerator of broader AI uptake. Deployment continues to be most visible in general purpose, language- and content-based tools including document drafting, summarisation, translation, coding support, internal search, knowledge management assistants and call centre or adviser support - while direct customer-facing or core financial decisioning applications remain more limited and more controlled. In Japan, among firms already using AI, more than 70% allow generative AI use broadly for general employees, and document summarisation, translation and proofreading have each already been implemented by more than 70% of surveyed financial institutions. In France, the AMF finds that generative AI is the most frequently reported AI technology among detailed financial-services use cases, accounting for 52% of declared technologies; the same survey shows that 83% of detailed AI use cases are internal, with drafting, summarisation and translation and internal co-pilots each representing 21% of use cases, while 17% relate to the client relationship and only 1% apply directly to investment services. In European insurance, EIOPA reports that 65% of undertakings are already using generative AI and a further 23% plan to do so within three years, although 64% of use cases remain at proof-of-concept or experimentation stage and 64% are back-office rather than customer-facing. Hong Kong?s HKMA similarly describes the sector as moving from traditional AI towards generative and agentic AI, while noting that most generative AI implementations remain internal and non-customer-facing.

Current deployments are concentrated in activities where AI can support staff, process information or automate defined workflows without replacing core financial decision-making. In France, 83% of the 106 detailed use cases reported to the AMF were internal, while 17% related to client relationships and 1% directly to investment services. The most common categories were drafting, summarisation and translation and internal co-pilots, each at 21%, followed by data quality and processing tools and code generation, each at 8%. Generative AI was the most frequently reported technology, at 52% of declared technologies. In Japan, document summarisation, translation and proofreading had each already been implemented by more than 70% of surveyed financial institutions. In Dutch asset management, AI is used mainly for information sourcing, unstructured data analysis and research drafting..

Firms report chatbots, voicebots, customer communication tools, call centre support, adviser support, email-response tools and claims or onboarding assistance, but many of these use cases remain staff-mediated or subject to human validation. In European insurance, EIOPA identified 957 generative AI use cases, of which 64% were back-office and 36% customer-facing, with 64% still at proof-of-concept or experimentation stage. Current insurance adoption is highest in customer service, claims management and sales and distribution, at 40%, 32% and 22% of undertakings respectively, while planned three-year adoption is highest in fraud detection at 64%, claims management at 59% and sales and distribution at 54%.

Reported applications include market research, portfolio and risk analytics, alternative data processing, sentiment analysis, trading algorithm optimisation, price forecasting and compliance surveillance. Japan identifies use cases in foreign exchange and interest rate forecasting, portfolio optimisation, strategy enhancement, real-time sentiment analysis using news and social media, and alternative data such as satellite imagery, geolocation, traffic data and supply-chain information. In Dutch asset management, proprietary traders stand out for more complex uses: 65% use AI to optimise trading- lgorithm parameters, 62% to improve trading strategies and 62% to predict price movements or market prices.

Current examples point to systems that can plan, monitor and execute multi-step tasks within defined controls, but adoption remains early. Observed use cases include claims triage and low-value claims handling, AML/KYC and source-of-wealth support, internal research agents, employee workflow assistants, call-centre support, personal financial management tools and agentic payments pilots. In insurance, EIOPA reports that 84 of 957 generative AI use cases were labelled agentic, split between 49 customer-facing and 35 back-office use cases. Examples include chatbots and voicebots, personalised advertising banners, call summarisation, automated processing and settlement of low-value claims, invoice assessment, automated email response, structured data generation from contracts and customer query intent recognition. In France, agentic AI systems accounted for 8% of reported AI technologies, with respondents noting that autonomy remains relatively limited. In Dutch asset management, 253 of 323 respondents, or 78%, did not plan to implement AI agents or multi-agent systems within 12 months.

Despite this upward trajectory, new insights also sharpened the diagnosis of the practical barriers that prevent AI from scaling.

Several studies point to the limitations of current adoption and challenge in evidencing measurable impact and value. The Hong Kong Monetary Authority, as part of its Fintech Blueprint, notes that many generative AI deployments remain internal and non-customer-facing, limiting measurable ROI, and highlights persistent scepticism where models cannot yet replace front-office expertise or are not trusted enough for revenue-sensitive use cases. The French AMF similar finds that most reported AI use cases among market actors remain internal and low impact. The Japan FSA notes that ROI can be difficult to demonstrate, particularly where benefits are indirect, long-term, usage-dependent, or tied to broader process transformation rather than immediate cost reduction.

Concerns around data readiness—ranging from poor data quality and inconsistent data formats to fragmented data architectures and legacy system integration issues—alongside challenges in maintaining retrieval-augmented generation knowledge bases, appear as recurring practical blockers. Respondents to a survey by the Dutch AFM cite data quality as the most critical implementation challenge, identified by 50% of respondents, followed by data protection at 42%.

As financial institutions rely on external models, commercial cloud infrastructure, AI vendors and data providers, procurement and contract management are becoming central elements of AI governance. Firms are concerned not only about vendor selection, but also about transparency, auditability, contractual access, model updates, data processing locations, switching risk, concentration and lock-in. This is particularly pronounced for generative AI: EIOPA finds that insurers commonly purchase off-the-shelf solutions or build on third-party models. France’s AMF finds heavy reliance on commercial third-party generative AI models and concentration among a small number of mostly non-European providers. The UK Treasury Committee on its report on AI adoption in financial services highlights reliance on a small number of US technology firms for AI and cloud services as an operational resilience concern.

Firms face practical uncertainty over what data can be used in AI systems, where that data may be processed or stored, and how responsibilities are allocated when models, cloud infrastructure or vendors sit outside the home jurisdiction. Japan’s FSA highlights uncertainty around using customer data in AI-enabled sales support, entering personal data into prompts, relying on overseas generative AI platforms, and outsourcing model development or training. Likewise, a joint study by Central Asian authorities identifies cross-border data exchange and confidentiality as key constraints to regional scaling.

Hong Kong reports a substantial skills gap: around 76% of institutions cite technical skills shortages in generative AI development and use, while 60% cite compliance skills shortages. The most acute gap is in hybrid expertise: professionals who can combine AI technical capability with financial regulation, model-risk management, data governance, cybersecurity, business process knowledge and local market operations experience. This is consistent with EIOPA’s finding that lack of skilled staff is one of the leading implementation barriers for generative AI in insurance.