The Convergence of Artificial Intelligence and Blockchain in Financial Systems: Opportunities, Challenges, and Future Directions

Abstract

Artificial Intelligence and blockchain are converging in ways that are quietly reshaping how financial systems operate. AI brings predictive analytics, automated decision-making, and fraud detection; blockchain contributes an immutable, decentralised record that can be independently verified. Taken together, applications such as AI-augmented smart contracts and blockchain-anchored data pipelines are already changing how fraud is detected, how compliance is handled, and how decentralised finance (DeFi) functions. But the same combination that makes these systems powerful also makes them harder to govern: technical, regulatory, and ethical obstacles still stand in the way of adoption at scale. This paper uses a targeted, purposive literature synthesis alongside exploratory case analysis of financial institutions and fintech platforms to examine how AI and blockchain are transforming finance together, what barriers and systemic risks accompany that transformation, and where research and regulation need to go next. Drawing on the Technology Acceptance Model, Diffusion of Innovation Theory, and the Socio-Technical Systems perspective, the paper builds a multi-level framework for thinking about how AI–blockchain convergence can be adopted responsibly across the financial industry.

1. Introduction

Innovation has always been central to how financial institutions compete, and the sector has a long history of adopting new tools to cut costs, improve transparency, and manage risk more effectively. Two of the more disruptive technologies to emerge in recent years, Artificial Intelligence (AI) and blockchain, have each already proven capable of reshaping how financial value gets created, recorded, and moved. What happens when the two are combined is arguably more interesting than either on its own: the convergence carries compounding upside and, just as importantly, compounding risk, which is why it deserves closer scholarly attention.

AI, in the broad sense that includes machine learning (ML), predictive analytics, natural language processing (NLP), and automated decision-making, already lets financial institutions sharpen credit scoring, run algorithmic trading, detect fraud, and serve customers at a scale that was simply not possible before [1,2]. Blockchain, meanwhile, has reshaped payments, settlement, asset tokenisation, and decentralised finance through its decentralised, immutable, and transparent ledger structure [3,4].

AI and blockchain have each been studied extensively on their own, but their intersection is still a relatively young area of research. There are real synergies worth exploring here: AI models can make sense of the large volumes of data that blockchain networks generate, while blockchain in turn gives AI something it often lacks—a verifiable, tamper-resistant source of truth [5]. Smart contracts enhanced with AI, for instance, can automate compliance workflows for Know Your Customer (KYC) and Anti-Money Laundering (AML) checks, and blockchain can reduce the risk of AI models being fed manipulated data by guaranteeing that inputs are verifiable [6].

None of this comes free of complications, though. On the technical side, blockchain’s scalability and interoperability limits make it hard to support real-time AI-driven financial applications [7]. On the regulatory side, governance frameworks for AI and blockchain are still fragmented across jurisdictions, leaving institutions uncertain about what large-scale adoption will actually require [8]. And on the ethical side, the opacity that already troubles AI models becomes a bigger problem once those models are locked into immutable blockchain systems, raising harder questions about accountability, bias, and privacy [9,10].

Most existing scholarship treats AI and blockchain as separate subjects, and the work that does look at their joint application in finance tends to stay narrowly focused on technical feasibility rather than what it means for the financial system as a whole [11,12]. This paper tries to close that gap by examining AI–blockchain convergence in finance in a more integrated way, guided by three research questions:

• How does the integration of AI and blockchain reshape financial systems with respect to efficiency, transparency, and risk management?
• What are the principal barriers, risks, and ethical concerns associated with adopting AI–blockchain solutions in finance?
• What directions should future research and regulatory frameworks pursue to enable sustainable adoption?

To answer these, the paper combines a targeted, purposive literature synthesis with exploratory case study analysis, drawing evidence from financial institutions and fintech platforms. The aim is to offer a theoretically grounded account of AI–blockchain convergence that is also useful in practice, for regulators, institutions, and technology developers alike.

2. Literature Review

Finance has always been an early adopter of new technology, using it to cut costs, improve efficiency, and build stakeholder trust. This section reviews what the literature says about blockchain and AI in finance separately, then turns to what happens when the two converge.

2.1. Blockchain in Finance

Blockchain started life as the infrastructure behind Bitcoin in 2008 and has since grown into a much broader platform for financial innovation [3]. At its core, it is a distributed ledger technology (DLT): a way of recording transactions that is secure, immutable, and decentralised [4]. Those three properties—trust, auditability, and security—are exactly what make it valuable for financial services [9,10]. Throughout this paper, we use “blockchain” and “distributed ledger technology (DLT)” largely interchangeably, following common usage in the finance and fintech literature. Strictly, blockchain is one specific implementation of the broader DLT category, distinguished by its sequential, cryptographically linked block structure; other DLT architectures (e.g., directed acyclic graphs) organise data differently. We note the distinction explicitly wherever it matters for a specific application discussed below.

One of its clearest use cases is in payments and cross-border settlement. Traditional cross-border transactions are slow, expensive, and pass through several intermediaries along the way. Blockchain allows near-instant peer-to-peer value transfer at a fraction of the cost, which could disintermediate existing settlement chains and open up remittance access for populations that banks have historically underserved [6,13].

A second major application is asset tokenisation and decentralised finance (DeFi). Real estate, equities, and commodities can be converted into digital tokens, which makes fractional ownership possible and improves liquidity [12]. DeFi platforms take this further still, building lending, borrowing, and trading mechanisms that operate without traditional intermediaries at all [5]. This expands financial inclusion, but it also introduces systemic instability risks that governance frameworks have yet to fully address.

A third area is fraud prevention and auditability. Because blockchain records cannot be altered after the fact, they close off many of the usual avenues for financial manipulation and make transactions easier to trace, particularly in supply-chain and trade finance [7]. Central banks have taken note too, exploring blockchain-inspired architectures for Central Bank Digital Currencies (CBDCs) as a way to strengthen monetary policy transmission and financial oversight [8].

That said, blockchain has real constraints. Throughput is one of the biggest: Bitcoin processes roughly seven transactions per second, nowhere near Visa’s capacity, which is a genuine scalability problem [4]. Proof-of-work consensus mechanisms also raise environmental concerns [14], and regulatory fragmentation complicates any cross-border deployment—Switzerland and Singapore have moved toward innovation-friendly frameworks, while other jurisdictions remain restrictive or simply undecided [5].

2.2. Artificial Intelligence in Finance

AI has found its way into finance largely because it is good at exactly what the sector needs: processing large volumes of structured and unstructured data, spotting complex patterns, and supporting decisions made under uncertainty [1]. Financial institutions now use it across almost every function, from risk modelling to customer engagement.

Algorithmic trading was one of the earliest and remains one of the most significant AI applications in finance. Machine learning models can read real-time market data and execute trades at speeds and volumes no human trader could match [2]. High-frequency trading firms and hedge funds increasingly rely on reinforcement learning and neural networks to refine trading strategies and capture arbitrage opportunities [11].

Credit risk assessment has also changed considerably. Traditional scoring systems lean heavily on historical credit bureau data, which tends to systematically exclude underbanked populations. AI-driven models widen the net by pulling in alternative data—transaction histories, utility payments, behavioural patterns—to improve access to credit and financial inclusion more broadly [4], which fits well with wider goals around sustainable and inclusive finance.

Fraud detection is another area where AI has had a large impact. Supervised and unsupervised machine learning algorithms can spot anomalies in transaction data and flag suspicious activity faster and more accurately than rule-based systems ever could [5]. Banks using AI-based anomaly detection can now monitor millions of daily transactions while cutting down both false positives and undetected fraud [15].

Robo-advisory services show a different side of AI’s impact: the democratisation of wealth management. These systems produce personalised, algorithm-driven investment recommendations based on a client’s risk appetite and goals at a fraction of what traditional advisory services cost [12]—and they have found particular traction among younger, digitally fluent investors.

None of this is without risk, though. Deep learning models are often opaque, making their decisions hard to explain or audit, which is a growing sticking point for regulators who expect explainability in consequential decisions like loan approvals or fraud classifications [1,14]. Models trained on historically biased data can also reproduce discriminatory outcomes in lending and insurance [2], and the data privacy and cybersecurity risks are heightened in finance, given how much sensitive personal and institutional data underpins model training [8].

2.3. AI–Blockchain Synergies in Finance

We define genuine AI–blockchain convergence as cases where the two technologies are functionally interdependent: blockchain-recorded data materially shapes an AI system’s decisions, an AI system’s outputs are executed or committed via blockchain infrastructure, or both, such that neither technology could deliver the observed capability alone. This is distinct from mere co-occurrence, in which an institution uses AI and blockchain for unrelated purposes within the same organisation. We apply this criterion explicitly when discussing our three cases in Section 4.3, Section 5.1, Section 5.2 and Section 5.3: Ant International’s Whale platform meets it directly (blockchain settlement and AI-supported processing are integrated in the same operational pipeline); Project AIKYA does not meet it in the strict sense, since it is a federated-learning system without a distributed-ledger component, and we accordingly describe it as convergence-adjacent rather than a convergence case; and MakerDAO meets the blockchain half of the criterion but not the AI half, on the evidence available to us, which is why we treat it as a boundary case throughout.

The intersection of AI and blockchain has been drawing more academic interest lately, though systematic study of it is still fairly recent. Individually, each technology has limitations, but together they complement each other well: AI supplies analytical intelligence, and blockchain supplies a trustworthy, tamper-resistant foundation for the data that intelligence works on [5]. The five synergy areas discussed below are not an exhaustive taxonomy; they reflect the most frequently recurring themes identified through the thematic coding of the review corpus (Table S1 in Supplementary Materials) and are illustrated with representative, rather than comprehensive, examples. Table 1 summarises them.

Smart contracts are one of the clearest points of convergence. AI can give smart contracts the ability to make adaptive, real-time decisions based on financial data as it comes in—an AI-enhanced smart contract, for example, could adjust a loan’s interest rate as a borrower’s risk profile changes, while blockchain guarantees that the contract’s execution stays immutable and enforceable [12].

Fraud detection and auditability form another important synergy. AI models are good at spotting suspicious transaction patterns, but their usefulness depends entirely on the integrity of the data feeding them. Blockchain supplies a tamper-evident audit trail that reduces the risk of that data being manipulated in ways that would compromise the AI’s inputs or outputs [13]. Together, the two technologies support a fraud detection architecture that is transparent, trustworthy, and fast. De Mariz [16] documents a related pattern across fintech more broadly: access to a financial technology is not sufficient on its own, and trust is the essential link between access and actual adoption—a dynamic also visible in emerging markets, which De Mariz shows have often leapfrogged directly to digital financial solutions rather than following the developed-market adoption path.

DeFi is where AI adds further value on top of existing blockchain infrastructure. DeFi platforms already run decentralised lending, borrowing, and trading; layering AI on top improves credit risk assessment, market trend prediction, and automated liquidity provision [7]. AI can also optimise portfolio allocation within decentralised asset management systems, effectively bringing institutional-grade analytics to individual users. Recent literature characterises smart contracts as a core infrastructural component of DeFi, while also identifying important limitations relating to external data dependencies, governance, legal enforceability, and implementation risk [17,18,19,20]. This mirrors the broader financial-inclusion argument in De Mariz [16], who shows that innovation of this kind is an essential tool for extending financial access while also posing new challenges for supervisory bodies and incumbent institutions.

Regulatory compliance—KYC and AML in particular—is a high-cost area where AI–blockchain convergence offers real efficiency gains. AI can handle identity verification and flag suspicious activity, while blockchain provides secure, transparent, user-controlled record-keeping that satisfies regulators’ evidentiary requirements [8]. Together, this lowers compliance costs while strengthening trust between institutions and regulators. Recent studies suggest that blockchain-supported identity infrastructures can reduce duplication in customer verification and facilitate more efficient KYC and AML processes, although implementation remains subject to regulatory, privacy, and interoperability constraints [21,22].

Privacy-preserving data architectures are an emerging area of convergence too. Blockchain handles decentralised data management and immutability, while AI techniques such as federated learning and homomorphic encryption allow models to be trained without ever directly accessing sensitive data—which helps with compliance under data protection regimes like GDPR and PIPEDA [14].

Table 1. AI–blockchain synergy areas in finance: illustrative examples (non-exhaustive).

Synergy Area	Representative Mechanism	Illustrative Example	Evidentiary/Convergence Status
Smart contracts	AI-driven adaptive contract terms executed on an immutable ledger	Dynamic loan-term adjustment based on real-time borrower risk data [12]	Illustrative mechanism; not evidence of a confirmed operational convergence case.
Fraud detection & auditability	ML anomaly detection combined with a tamper-evident audit trail	Cross-institutional anomaly detection without pooling raw data, e.g., Project AIKYA [23]	Convergence-adjacent: AI/federated-learning proof-of-concept; Project AIKYA itself does not contain a DLT component.
Decentralised finance (DeFi)	AI-supported credit risk assessment and liquidity optimisation on blockchain rails	Algorithmic collateral and governance mechanisms in protocols such as MakerDAO [24]	Boundary example: blockchain/DeFi case; no verified AI component in the evidence examined.
Regulatory compliance (KYC/AML)	AI-driven identity verification paired with blockchain-secured records	Blockchain-based KYC architecture reducing duplicated verification steps [21]	Research-based illustrative application; not presented here as a confirmed operational AI–blockchain deployment.
Privacy-preserving data sharing	Federated learning and homomorphic encryption over decentralised data	Multi-jurisdictional model training without centralising sensitive records [14]	Illustrative/research-based mechanism; operational AI–blockchain convergence is not independently established by this example.

Note: Examples are classified according to the functional-interdependence criterion defined in Section 2.3. Convergence-adjacent and boundary examples are retained deliberately to distinguish confirmed AI–blockchain convergence from technological co-occurrence, related decentralised-AI architectures, and blockchain-only adoption.

Table 1. AI–blockchain synergy areas in finance: illustrative examples (non-exhaustive).

Synergy Area	Representative Mechanism	Illustrative Example	Evidentiary/Convergence Status
Smart contracts	AI-driven adaptive contract terms executed on an immutable ledger	Dynamic loan-term adjustment based on real-time borrower risk data [12]	Illustrative mechanism; not evidence of a confirmed operational convergence case.
Fraud detection & auditability	ML anomaly detection combined with a tamper-evident audit trail	Cross-institutional anomaly detection without pooling raw data, e.g., Project AIKYA [23]	Convergence-adjacent: AI/federated-learning proof-of-concept; Project AIKYA itself does not contain a DLT component.
Decentralised finance (DeFi)	AI-supported credit risk assessment and liquidity optimisation on blockchain rails	Algorithmic collateral and governance mechanisms in protocols such as MakerDAO [24]	Boundary example: blockchain/DeFi case; no verified AI component in the evidence examined.
Regulatory compliance (KYC/AML)	AI-driven identity verification paired with blockchain-secured records	Blockchain-based KYC architecture reducing duplicated verification steps [21]	Research-based illustrative application; not presented here as a confirmed operational AI–blockchain deployment.
Privacy-preserving data sharing	Federated learning and homomorphic encryption over decentralised data	Multi-jurisdictional model training without centralising sensitive records [14]	Illustrative/research-based mechanism; operational AI–blockchain convergence is not independently established by this example.

Note: Examples are classified according to the functional-interdependence criterion defined in Section 2.3. Convergence-adjacent and boundary examples are retained deliberately to distinguish confirmed AI–blockchain convergence from technological co-occurrence, related decentralised-AI architectures, and blockchain-only adoption.

2.4. Research Gaps

The volume of systematic and bibliometric review activity in this space has grown substantially over the past three years [25,26,27,28], yet even with this growing interest, the literature still has some significant gaps.

First, most existing work is conceptual or technical rather than empirically validated. Case studies documenting successful, at-scale deployments of AI–blockchain systems inside operational financial institutions remain relatively rare [11].

Second, the regulatory dimension is underexplored. Fintech regulation and blockchain governance have each drawn substantial scholarly attention on their own [8,14], but questions of liability, accountability, and cross-border enforcement specific to AI–blockchain-driven finance are still not well addressed.

Relative to other recent systematic reviews in this space—which synthesise the AI-in-finance or blockchain-in-finance literatures at a conceptual level (e.g., refs. [22,27,28,29]—this paper’s distinctive contribution is to test whether claimed instances of AI–blockchain convergence hold up under case-level scrutiny, rather than treating reported convergence claims at face value. Case-level source verification further shows that not all examples commonly presented as AI–blockchain convergence satisfy a strict functional-interdependence criterion. Two of the three cases examined in this study do not fully meet this criterion and are therefore retained as convergence-adjacent or boundary cases. This distinction constitutes a substantive finding, demonstrating the importance of differentiating confirmed convergence from technological co-occurrence or blockchain-only adoption. We are not aware of another study in this literature that combines a targeted, structured literature synthesis with source-verified, evidentiary-tier classification of case evidence in this way.

Third, ethics and fairness have not received enough scrutiny. Blockchain’s transparency does not automatically fix AI bias—if anything, embedding a biased model into an immutable ledger risks locking discriminatory practices in place in ways that are difficult, or even impossible, to undo later [1]. Recent explainable AI (XAI) scholarship reinforces this concern, highlighting that trust, accountability, and regulatory acceptance of financial AI systems depend on interpretability safeguards that blockchain immutability alone does not provide [30].

Fourth, scalability and interoperability challenges still lack robust solutions in the literature. Blockchain architectures continue to struggle with throughput demands, while AI needs high-speed processing and large-scale data integration to function well [7,29].

Finally, long-term systemic implications remain underanalysed. Most studies focus on operational efficiency gains and pay less attention to how AI–blockchain convergence might affect systemic risk, financial stability, or monetary policy transmission—questions that matter more each year as central banks explore CBDCs and AI-driven regulatory tools.

3. Theoretical Framework

This paper draws on three complementary theories to make sense of AI–blockchain convergence in finance: the Technology Acceptance Model (TAM), Diffusion of Innovation (DoI) Theory, and the Socio-Technical Systems (STS) perspective. Together they cover three different levels of analysis—individual adoption, industry-wide diffusion, and broader socio-institutional integration—which is why using all three, rather than one, makes sense here.

3.1. Technology Acceptance Model (TAM)

Davis’s [31] Technology Acceptance Model holds that adoption intentions come down mainly to two things: perceived usefulness (PU) and perceived ease of use (PEOU). Applied to AI–blockchain convergence in finance, TAM suggests that whether institutions and individuals adopt these systems will depend on whether they see real value in them—better transparency, better fraud detection, more efficient compliance—and on whether implementing them feels manageable given existing infrastructure.

Blockchain-anchored AI credit scoring is a good example: it can dramatically improve data provenance and trust, but if integrating it with legacy banking systems feels too technically burdensome, adoption will stall regardless of how useful the system is in principle. TAM is therefore most useful here as a micro-level lens on what drives—or blocks—organisational and individual acceptance of AI–blockchain solutions.

3.2. Diffusion of Innovation (DoI) Theory

Rogers’s [32] Diffusion of Innovation Theory explains how innovations spread through social systems over time, shaped by five factors: relative advantage, compatibility, complexity, trialability, and observability. Applied to AI–blockchain integration in finance, these factors help explain several patterns in adoption:

−

Relative advantage: how much AI–blockchain systems outperform existing centralised approaches—for example, in cross-border payment speed or fraud detection accuracy.

−

Compatibility: how well these systems fit with existing regulatory frameworks, banking processes, and institutional technology infrastructure.

−

Complexity: how technically difficult integration is, which can slow adoption particularly at institutions with limited fintech capacity.

−

Trialability: whether low-risk pilot environments exist, such as CBDC sandboxes or blockchain-based KYC trials, that let institutions experiment before committing.

−

Observability: how visible successful adoption by peer institutions is, which tends to accelerate diffusion across the industry.

DoI is especially useful for understanding diffusion dynamics at the industry level—the meso level—and the roles that regulators, industry consortia, and market leaders play in shaping how adoption unfolds.

3.3. Socio-Technical Systems (STS) Perspective

Where TAM and DoI focus on what drives adoption, the Socio-Technical Systems perspective [33] places technological change within its broader human, organisational, and institutional context. STS theory’s central claim is that successful adoption requires the technical subsystem (algorithms, infrastructure, tools) and the social subsystem (regulatory bodies, institutional norms, cultural expectations) to evolve together, not separately.

For AI–blockchain convergence specifically, STS points to a few requirements for effective adoption: regulatory frameworks that keep pace with data privacy, cybersecurity, and financial governance concerns [14]; organisational restructuring of compliance workflows and IT architecture; and genuine cultural acceptance, from customers and stakeholders alike, of financial decisions being made by algorithms. STS also draws attention to unintended consequences—systemic risk becoming entrenched, labour being displaced, or algorithmic bias becoming permanently baked into immutable blockchain systems.

3.4. Integrative Framework

Together, TAM, DoI, and STS give this study a multi-level foundation: TAM covers micro-level individual and organisational adoption drivers, DoI covers meso-level industry diffusion, and STS covers macro-level socio-technical integration across regulatory, cultural, and ethical dimensions. Using all three keeps the analysis from focusing on technical feasibility alone—it also has to account for the human, organisational, and policy conditions that responsible adoption actually requires.

4. Methodology

4.1. Research Design

This study uses a qualitative, exploratory design that combines a targeted, purposive literature synthesis with case study analysis. Two things justify pairing them: academic understanding of AI–blockchain convergence in finance is still young and scattered, which calls for structured synthesis, and case studies give theoretical propositions something to stand on by showing how these systems actually play out across different institutional settings. The literature component employed a targeted and purposive literature-synthesis approach designed to identify recent and substantively relevant scholarship at the intersection of Artificial Intelligence, blockchain/distributed ledger technologies, and financial services. The analytical corpus was developed through targeted database searching, relevance-based screening, citation chaining, and source verification, followed by structured thematic coding. Because complete database-specific search logs and intermediate screening records were not retained, the review is not presented as a fully reproducible PRISMA-based systematic literature review. The TAM, DoI, and STS frameworks were applied to each case using a structured protocol: for each case, we first classified its evidentiary tier (confirmed operational deployment, pre-production proof-of-concept, or peer-reviewed research without a confirmed AI component); we then evaluated each framework’s constructs (e.g., perceived usefulness and ease of use for TAM; relative advantage, compatibility, and trialability for DoI; technical- and social-subsystem maturity for STS) against the primary and secondary source material specific to that case, rather than inferring them generically from the literature; and judgments were cross-checked against the original source documents (company disclosures, technical reports, or peer-reviewed studies) cited alongside each claim in Section 5.7.

4.2. Literature Synthesis

Rather than retrospectively reconstructing unverifiable database retrieval and screening statistics, the review is described transparently here as a documented analytical corpus assembled through targeted searching, citation-based identification, relevance screening, and verification, using predefined search terms, eligibility criteria, study-level documentation, and thematic coding to strengthen transparency and traceability.

4.2.1. Data Sources

The review drew on major academic databases—Scopus, Web of Science, IEEE Xplore, ScienceDirect, and Emerald Insight—along with Google Scholar for grey literature.

4.2.2. Search Approach

Targeted searches used combinations of terms relating to Artificial Intelligence or machine learning, blockchain or distributed ledger technology, and finance, banking, or fintech. Because complete database-specific search strings and search logs were not retained, these terms are reported as the conceptual search structure rather than as a fully reproducible Boolean protocol.

4.2.3. Inclusion and Exclusion Criteria

Included: peer-reviewed journal articles, conference papers, and book chapters that address AI–blockchain convergence in finance. The search considered literature published between 2010 and 2026 to capture the development of AI–blockchain convergence over time; the final documented review corpus used for thematic coding, however, comprises 28 peer-reviewed studies published between 2022 and 2026, for which complete bibliographic and coding information is available. Earlier seminal literature (e.g., refs. [3,31,32,33]) was retained separately where necessary to establish foundational theoretical and technological concepts. Excluded: purely technical studies with no financial relevance, non-English publications, and duplicates.

4.2.4. Study Selection

Potentially relevant studies identified through targeted database searching and citation-based identification were assessed for relevance against the stated eligibility criteria. The final documented analytical corpus comprises 28 peer-reviewed studies published between 2022 and 2026 for which complete bibliographic and thematic-coding information is available. Full bibliographic information and thematic coding for these studies are provided in Supplementary Table S1. This corpus includes [34,35,36,37,38,39,40,41,42,43,44,45,46,47], in addition to the sources cited throughout this paper.

Each entry in Supplementary Table S1 is linked to one or more theme codes (fraud detection, regulatory compliance, smart contracts, DeFi, and systemic risk management) matching the thematic categories used in the synthesis below, so that the coding decisions underlying Section 5 can be traced back to specific, identifiable sources.

4.2.5. Data Extraction

A structured extraction matrix was used to record bibliographic information, research objectives, study design, principal findings, and relevance to AI-DLT applications in finance. The included studies were coded using five predefined thematic categories: T1 = fraud/anomaly detection and risk management; T2 = compliance, KYC/AML, and explainability; T3 = smart contracts and programmable finance; T4 = DeFi, tokenisation, and decentralised governance; and T5 = systemic, privacy, interoperability, and governance risk. The thematic coding was conducted by the lead author using these predefined categories. Coding decisions were subsequently checked for internal consistency across the analytical corpus. As independent duplicate coding was not undertaken, an inter-coder reliability coefficient was not calculated; this is acknowledged as a limitation in Section 6.5.

4.2.6. Bibliometric Profile of the Review Corpus

A brief bibliometric profile of the review corpus (Supplementary Table S1) illustrates its composition and recency. Of the 28 included studies, 2 were published in 2022, 9 in 2023, 10 in 2024, 6 in 2025, and 1 in early 2026, reflecting the rapid recent growth in AI–blockchain scholarship noted in Section 2.4. Across the five theme codes, systemic, privacy, and governance risk (T5) is the most broadly represented, appearing in 26 of the 28 studies, followed by compliance, KYC/AML, and explainability (T2, 17 studies) and DeFi, tokenisation, and decentralised governance (T4, 15 studies); fraud and anomaly detection (T1, 12 studies) and smart contracts (T3, 11 studies) are comparatively less represented in the corpus. This distribution suggests that governance and systemic-risk considerations currently receive more sustained scholarly attention than operational AI applications such as fraud detection, and it informed the selection of the three cases in Section 4.3, each chosen to probe one or more of the more heavily represented themes.

4.3. Case Study Analysis

Exploratory case studies were added to ground the literature synthesis’s theoretical insights in actual institutional practice. Cases were chosen through purposive sampling based on relevance to themes identified in the review, variety across institution types, and sufficient data accessibility through company reports, technical documentation, and peer-reviewed sources. The three cases selected were: Project AIKYA, a federated-learning proof-of-concept developed by Kinexys for J.P. Morgan and BNY (anomaly detection); Ant International’s Whale platform, a blockchain-AI treasury settlement infrastructure; and MakerDAO (DeFi governance and oracle risk). The cases are illustrative rather than representative, and are not intended to causally validate the literature synthesis’s thematic findings; they were purposively selected to probe the boundaries of the AI–blockchain convergence construct itself. As detailed in Section 5.1, Section 5.2 and Section 5.3, the strength and type of evidence vary considerably across the three: Ant International’s Whale platform is a documented, company-reported operational deployment; Project AIKYA is an explicitly experimental, non-production proof-of-concept using synthetic data; and MakerDAO is supported by peer-reviewed governance research but does not, on the evidence available to us, involve a documented AI component. Data for each case was triangulated across company disclosures, technical documentation, and peer-reviewed sources, and each case’s evidentiary basis is stated explicitly rather than assumed uniform across cases.

4.4. Validity and Reliability

Methodological consistency was supported through predefined eligibility criteria, a structured data-extraction framework, and explicit thematic coding categories. The complete set of studies included in the thematic synthesis and their assigned theme codes are reported in Supplementary Table S1, allowing the analytical pathway from source selection to thematic interpretation to be traced.

This study relies entirely on secondary data—academic publications and publicly available case material. Attribution was handled carefully throughout, and findings were represented as objectively as possible when reporting on institutional practices and regulatory frameworks. Because the analysis relies entirely on secondary sources, quantitative or operational outcomes reported in the case discussion are attributed to their original sources and should be interpreted according to the evidentiary status of those sources rather than as independently verified causal effects.

5. Findings and Discussion

5.1. Fraud Detection and Risk Management: Evidence from a Proof-of-Concept

One of the most consistent themes in the literature is AI–blockchain convergence being applied to fraud prevention and risk management [48,49]. Machine learning is well suited to spotting transaction anomalies, and blockchain keeps the underlying data immutable, transparent, and resistant to tampering [13].

Project AIKYA, a collaboration between Kinexys by J.P. Morgan and BNY, illustrates a convergence-adjacent proof-of-concept in this space rather than an operational deployment. In a federated-learning framework, participating institutions train anomaly-detection models locally and share only aggregated model updates, avoiding the need to pool raw transaction data. The project reports that a globally aggregated model performed as well as, or better than, individually trained models, and that federated aggregation improved detection coverage by combining institution-specific patterns [23]. Critically, the proof-of-concept uses synthetic, not real, transaction data, and its developers explicitly describe it as not production-ready and not validated for operational, commercial, or compliance-sensitive use. AIKYA is not itself a distributed-ledger application; we treat it here as convergence-adjacent because it operationalises a decentralised-trust logic—model intelligence without centralised data pooling—conceptually related to the AI–blockchain convergence this paper examines, without itself being a blockchain case.

The broader takeaway is more modest than efficiency-focused literature in this space often suggests: decentralised-AI approaches show experimental promise for institutional fraud detection, but the evidentiary base remains at the proof-of-concept stage, with real-data validation, multi-participant scaling, and regulatory acceptance still ahead [23].

5.2. Blockchain-AI Convergence in Treasury Settlement and Compliance

A second major theme is the use of AI to make smart contracts more capable. Blockchain already provides immutability and automated execution; AI adds the adaptive intelligence needed for dynamic decision-making on top of that. AI can adjust loan terms as a borrower’s risk profile changes, for instance, while blockchain ensures the revised terms get executed without any human intervening [12].

Ant International’s Whale platform offers a documented, company-reported illustration of blockchain-AI convergence in institutional treasury operations. Whale is a proprietary treasury-management infrastructure that combines blockchain technology, advanced encryption (including homomorphic encryption and zero-knowledge proofs), and AI to move funds between Ant International’s own entities in real time. According to Ant International’s 2024 Sustainability Report, more than a third of the company’s transactions were processed on-chain via Whale in 2024, and the platform has since been extended to external bank partners—including Standard Chartered, HSBC, DBS, and OCBC—for cross-border settlement [50]. This is best characterised as a large-scale, company-reported operational deployment rather than an independently audited academic case study; the efficiency and settlement figures come from the company’s own disclosures. Separately, blockchain-based KYC architectures have been documented in banking more generally [21], illustrating a related but distinct application of blockchain-AI convergence to compliance functions specifically, in contrast to Whale’s treasury-settlement focus.

Real challenges remain. Legal liability when AI-augmented financial infrastructure produces erroneous or disputed outcomes is still not well resolved under current regulatory frameworks, and Whale’s evidentiary basis rests on company disclosure rather than independent audit. Working out accountability mechanisms for automated treasury and settlement decisions of this kind is a priority that both scholars and policymakers need to take seriously.

5.3. Decentralised Finance (DeFi) and Blockchain Governance Risk

DeFi is one of the most significant and fastest-moving applications of AI–blockchain convergence. DeFi platforms use blockchain infrastructure to offer lending, trading, and asset management without going through traditional intermediaries, and AI adds improved credit risk assessment, liquidity optimisation, and algorithmic governance on top [7]. The distinction between centralised and decentralised financial infrastructures, and the institutional risk implications of each, has also received renewed attention in the recent literature [51].

MakerDAO is best treated as a boundary case for this paper’s framework rather than a clear instance of AI–blockchain convergence. Peer-reviewed research documents MakerDAO’s blockchain-based governance mechanics in detail, including evidence of governance centralisation and concentrated voting power among a small number of large token holders [24], as well as the protocol’s dependence on price oracles and its exposure to collateral-liquidation risk during periods of market stress. We do not find independently verifiable evidence that MakerDAO deploys AI-driven collateral risk prediction; the protocol’s risk parameters are set through decentralised governance votes rather than a documented AI system. We retain MakerDAO in the analysis because it illustrates the limits of blockchain-only decentralisation—governance concentration and oracle dependence—that AI–blockchain convergence is often proposed to address, even where AI itself is not demonstrably present in the protocol’s operation.

The risks that are well documented here are governance-related rather than specifically algorithmic: concentrated voting power creates the potential for self-serving governance decisions, oracle failures can trigger cascading liquidations, and current regulatory frameworks are not well equipped to supervise decentralised, token-holder-governed protocols of this kind [24].

Table 2. Comparative analysis of case studies.

Case	Institution Type	AI Application	Blockchain Application	Key Outcomes	Challenges
Project AIKYA (Kinexys/J.P. Morgan & BNY)	Bank collaboration (proof-of-concept)	Federated anomaly-detection modelling (experimental)	Not a distributed-ledger application; decentralised-AI approach conceptually adjacent to blockchain convergence	Improved detection coverage in experimental testing using synthetic data; not yet validated on real data	Pre-production status; no real-data validation; not yet tested at multi-participant scale
Ant International (Whale platform)	Fintech/financial infrastructure	AI-supported transaction processing and encryption	Blockchain-based treasury settlement ledger (Whale)	Over one-third of Ant International’s transactions processed on-chain in 2024; expanding bank partnerships	Evidence is company-reported; independent audit not available; cross-jurisdictional regulatory complexity
MakerDAO (DeFi)	Decentralised platform	No independently verified AI component	Stablecoin governance via smart contracts; oracle-fed pricing	Documented governance and liquidity mechanisms; peer-reviewed evidence of governance centralisation	Voting-power concentration; oracle dependency and liquidation risk; regulatory uncertainty

Source: Compiled by authors.

summarises the three cases across institution type, AI and blockchain application, key outcomes, and evidentiary challenges.

5.4. Data Privacy, Security, and Trust

A fourth major finding is about how AI–blockchain convergence handles data privacy. Blockchain’s immutability and decentralised control, combined with AI privacy-preserving techniques like federated learning and homomorphic encryption, make it possible to train AI models on distributed financial data without ever centralising the sensitive information itself [14].

This matters especially for cross-border finance, where differing data protection rules—the EU’s GDPR and Canada’s PIPEDA, for example—make sharing information across jurisdictions genuinely complicated. Privacy-preserving AI–blockchain architectures offer one route toward regulatory compliance and stakeholder trust without giving up analytical usefulness.

Still, real tensions remain. Blockchain’s transparency is useful for auditability, but it can clash with individual privacy rights—the GDPR’s right to erasure, for instance, is technically incompatible with a public blockchain ledger’s immutability. Embedding opaque AI models within records that can never be changed also raises real accountability concerns. These tensions point to the need for governance structures that can genuinely balance transparency, privacy, and explainability rather than treating them as separate problems.

5.5. Systemic Risks and Regulatory Challenges

The literature and the case evidence both point to growing concern about what AI–blockchain adoption looks like at scale. Efficiency and transparency benefits are well documented at this point, but fewer studies actually examine the macro-financial consequences of widespread deployment [8].

At the macroeconomic level, widespread AI-driven DeFi could disintermediate traditional banking and weaken central banks’ ability to transmit monetary policy through the usual channels. At the systemic level, correlated algorithmic failures across interconnected blockchain ecosystems could amplify financial shocks rather than absorb them. The STS framework helps explain why technological adoption can outpace governance capacity when the social, legal, and institutional structures around it are not evolving at the same speed. The mechanism for amplification is structural rather than purely behavioural: many DeFi and blockchain-based platforms draw on a comparatively small number of shared AI risk models, price oracles, and liquidity pools, so a flaw or adversarial manipulation affecting one input can propagate simultaneously across otherwise-independent platforms. Blockchain composability compounds this risk, since smart contracts on different platforms often call one another automatically; a mispriced or manipulated input can trigger cascading liquidations across interconnected protocols without the human circuit breakers that exist in traditional financial infrastructure, producing a dynamic broadly analogous to correlated algorithmic behaviour in traditional markets (e.g., the 2010 equity flash crash) but with fewer institutional safeguards against it.

Existing regulatory models were built for centralised financial intermediaries, and they are not well suited to decentralised, algorithm-governed systems. Questions of jurisdictional authority, legal accountability, and cross-border enforcement in AI–blockchain networks remain largely unresolved—arguably the single biggest priority for regulatory scholarship and policy development going forward.

5.6. Synthesis of Findings

Taken together, the findings suggest that AI–blockchain convergence in finance carries both real opportunity and real challenge:

• Opportunities: reduced fraud, cost savings, democratisation of financial services, stronger regulatory compliance, and improved data security.
• Challenges: technical scalability constraints, ethical risks from bias and opacity, systemic financial vulnerabilities, and gaps in regulatory governance.

Viewed through the TAM-DoI-STS framework, it becomes clear that technical innovation alone will not be enough to realise what AI–blockchain systems are capable of. Organisational readiness, regulatory harmonisation, and building cultural trust all matter just as much for sustainable adoption.

Table 3. AI–blockchain integration benefits and challenges.

Dimension	Benefits	Challenges
Efficiency	Faster settlements; automated decision pipelines	Blockchain scalability constraints; legacy integration complexity
Transparency	Immutable, tamper-evident audit trail for regulators	Tension between on-chain transparency and individual data privacy
Financial Inclusion	DeFi expands access to underserved and unbanked populations	Algorithmic bias may replicate or amplify existing financial exclusions
Compliance	Automated KYC/AML reduces operational burden	Fragmented, jurisdiction-specific regulatory requirements
Risk Management	Real-time fraud detection; predictive analytics at scale	Systemic risk from algorithmic failures; governance accountability gaps

Source: Compiled by authors.

summarizes these opportunities and challenges.

5.7. Theoretical Synthesis: TAM, DoI, and STS in Light of the Case Evidence

Building on the case analyses in Section 5.1, Section 5.2 and Section 5.3, this section offers a brief theoretical synthesis rather than an independent empirical test. TAM, DoI, and STS are used here as interpretive lenses to make sense of why the three cases differ so sharply in their adoption trajectories, given that they occupy three different evidentiary tiers: confirmed operational deployment (Ant International’s Whale platform), a pre-production proof-of-concept (Project AIKYA), and peer-reviewed governance research without a confirmed AI component (MakerDAO).

Table 4. Cross-case theoretical mapping—TAM, DoI, and STS dimensions across three cases.

Theoretical Dimension	Project AIKYA (Kinexys/J.P. Morgan & BNY)	Ant International (Whale Platform)	MakerDAO (DeFi)
TAM (Perceived Usefulness/Ease of Use)	Not yet assessable: pre-production, propositional evidence only [23].	High and demonstrated at scale; company-reported [50].	Useful to sophisticated liquidity providers for governance mechanics; no confirmed AI component to evaluate.
DoI (Relative Advantage, Trialability, Observability)	Advantage and observability remain propositional; trialability applied deliberately via a bounded, synthetic-data design.	High relative advantage and observability; capability extended incrementally via bank partnerships.	Advantage attributable to blockchain mechanics rather than AI; low trialability given live deployment; technically but not practically observable [24].
STS (Technical–Social Co-Evolution)	Not yet testable; the technical subsystem exists only as a controlled experiment.	Comparatively mature; includes participation in regulatory sandboxes (e.g., Hong Kong’s Project Ensemble).	Genuine co-evolution deficit: a mature technical subsystem operates alongside a minimal, concentrated social subsystem [24].

Source: Authors’ analysis based on case documentation and theoretical frameworks [31,32,33].

summarises the key pattern for each framework across the three cases.

TAM’s usefulness/ease-of-use logic applies most straightforwardly to Ant International’s Whale platform, where both dimensions are supported by company-reported, at-scale evidence [50]. For Project AIKYA, TAM can only be applied propositionally, since the project has not been deployed for institutional use [23]. MakerDAO complicates TAM differently: its blockchain-governance mechanics are useful to sophisticated liquidity providers, but the case does not, on available evidence, involve the AI component TAM would need to evaluate. Taken together, the three cases suggest that TAM cannot be meaningfully applied to AI–blockchain convergence without first establishing which evidentiary tier a case occupies.

Rogers’s [32] DoI attributes help explain the diverging trajectories: relative advantage and observability are clearest for Whale, whose real-time settlement and transaction volumes are publicly documented, while both remain propositional for AIKYA and are technically but not practically observable for MakerDAO [24]. Trialability is the most theoretically interesting dimension: AIKYA’s bounded, synthetic-data design is trialability applied deliberately, whereas MakerDAO’s direct deployment to live blockchain infrastructure leaves little room for low-consequence trialling—a structural limitation in Rogers’s framework when applied to irreversible-deployment technologies.

The Socio-Technical Systems perspective has the most explanatory power here, since it foregrounds a co-evolution deficit that differs by case rather than following a uniform pattern: Whale shows comparatively mature co-evolution, including participation in regulatory sandboxes such as Hong Kong’s Project Ensemble; AIKYA’s co-evolution has not yet been tested, since the project remains pre-deployment; and MakerDAO shows a genuine deficit, with a mature technical subsystem operating alongside a minimal, concentrated social subsystem [24]. Blockchain immutability also emerges as a double-edged property in the one case where it is clearly present (Whale): it supports auditability, but the same immutability complicates correcting errors after the fact—a design trade-off rather than an unconditional benefit.

Reading the three frameworks together surfaces cross-cutting tensions, summarised in

Table 5. Cross-framework tensions revealed by case analysis—theoretical and policy implications.

Tension	What the Cross-Case Analysis Reveals	Implication for Theory or Policy
Evidentiary Tier Heterogeneity	TAM, DoI, and STS all implicitly assume an adoption event has occurred and can be explained. Across our three cases, only Ant International’s Whale platform clearly satisfies that assumption; Project AIKYA is pre-adoption, and MakerDAO demonstrates blockchain governance adoption without a confirmed AI component.	Studies of AI–blockchain convergence should classify cases by evidentiary tier (confirmed deployment, pre-production PoC, adjacent-technology adoption) before applying adoption-theory constructs comparatively across them.
DoI Trialability Gap in DeFi	Rogers’s [32] DoI model assumes innovations can be trialled before full commitment. MakerDAO’s deployment to live blockchain infrastructure violates that assumption; Project AIKYA’s bounded, synthetic-data design shows what a deliberate response to this gap can look like.	Future research should develop a modified DoI framework for irreversible-deployment technologies, replacing trialability with a risk-bounded experimentation dimension.
STS Co-Evolution Deficit	STS theory posits that technical and social subsystems must co-evolve for successful integration. This deficit is clearest and most consequential in MakerDAO, where a mature technical subsystem operates with a minimal, concentrated social subsystem; it is not yet testable for AIKYA and is comparatively mature for Whale.	Policymakers should prioritise closing the co-evolution gap in decentralised governance structures specifically, rather than treating AI–blockchain convergence as a uniform regulatory category.
Immutability as a Double-Edged STS Property	Blockchain immutability, often treated in the literature as an unqualified benefit, is a documented design trade-off in the one case where it is clearly present: Ant International’s Whale platform, where immutable settlement records aid auditability but complicate correcting errors or disputes after the fact.	System design should incorporate mechanisms for governed mutability—allowing authorised corrections under multi-party consensus—rather than treating immutability as an unconditional benefit.

Source: Authors’ analysis.

.

The most important tension, in our view, is evidentiary: TAM, DoI, and STS all implicitly assume an adoption event has occurred and can be explained, yet only the Whale case clearly satisfies that assumption. We suggest that studies of this kind classify cases by evidentiary tier before applying adoption-theory constructs comparatively across them, and that future work develop a modified DoI construct—replacing trialability with a risk-bounded experimentation dimension—for irreversible-deployment technologies such as MakerDAO’s.

Returning to the paper’s three research questions, this theoretical synthesis supports a qualified answer throughout: efficiency, transparency, and risk-management gains are demonstrated for Whale, propositional for AIKYA, and attributable to blockchain governance rather than AI for MakerDAO; the principal barriers differ by evidentiary tier rather than arising uniformly from AI opacity or bias; and future research and regulatory attention should prioritise evidentiary classification and the co-evolution deficit concentrated in decentralised governance structures, rather than treating AI–blockchain convergence as a single uniform category.

6. Conclusions and Implications

6.1. Key Findings

This paper set out to systematically examine AI–blockchain convergence in financial services, drawing on a structured literature review and three exploratory case studies. Five main findings emerge from that work.

First, the recent literature indicates that combining AI-enabled analytics with blockchain infrastructures has the potential to strengthen fraud detection, data integrity, auditability, and risk-management processes in financial services [5,13,22,29]. Second, the literature suggests that AI-augmented smart contracts may support more efficient KYC and AML processes while enabling adaptive financial decision-making [12]. Third, the literature indicates that AI capabilities may complement blockchain-based DeFi infrastructures through applications such as liquidity management and risk assessment [7]. Fourth, AI-enabled privacy techniques—federated learning especially—combined with blockchain’s immutable records, offer a promising path toward data protection compliance across multi-jurisdictional financial operations [14]. Fifth, none of these benefits come free: technical scalability constraints, ethical risks of algorithmic bias, systemic vulnerabilities in decentralised governance, and unresolved questions of regulatory accountability accompany them all [8].

6.2. Theoretical Contributions

This study makes three main theoretical contributions. It extends TAM into the domain of integrated AI–blockchain solutions in organisational finance, highlighting perceived usefulness and ease of use as the key adoption determinants. It applies DoI Theory to explain industry-wide diffusion dynamics, including the critical roles of relative advantage, compatibility, and observability. It uses the STS perspective to place technology adoption within its regulatory, organisational, and ethical context, giving the study a holistic, multi-level analytical foundation. Bringing these three frameworks together advances both the theoretical understanding and the practical analysis of this fast-moving domain.

6.3. Practical and Managerial Implications

For financial institutions, this study suggests piloting AI–blockchain solutions in controlled environments before committing to full-scale deployment, investing in staff training and legacy system integration to reduce adoption friction, and working proactively with regulators to build compliant, transparent financial services architecture.

For regulators, the findings point to a need for harmonised frameworks that accommodate AI–blockchain systems while still safeguarding financial stability, clear accountability mechanisms for algorithmic decisions embedded in immutable ledgers, and stronger promotion of transparency, explainability, and ethical design in financial AI applications.

For investors and technology developers, the study points to opportunities in building scalable, interoperable, and ethically designed AI–blockchain platforms that address emerging market needs around financial inclusion, risk management, and real-time regulatory compliance.

6.4. Policy Recommendations

Three specific policy recommendations follow from this analysis. First, governments and central banks should expand regulatory sandbox environments to allow structured experimentation with AI–blockchain systems under monitored conditions. Second, international bodies and standard-setting organisations should develop technical and ethical standards for AI–blockchain integration to support interoperability and institutional trust. Third, policymakers should build dedicated frameworks for monitoring the systemic risks that arise from algorithmic governance in decentralised financial platforms—an area current macroprudential toolkits do not really cover. On the specific question of accountability for biassed AI decisions recorded on immutable ledgers, we suggest three concrete mechanisms regulators could require: governed mutability layers that allow authorised correction of erroneous or biassed records under multi-party consensus, rather than treating immutability as absolute; staged commitment architectures that hold AI-generated decisions in a reviewable, off-chain state for a defined window before final on-chain commitment; and mandatory pre-deployment explainability and bias-testing certification for AI systems whose outputs will be committed to immutable infrastructure in consequential financial decisions. These mechanisms directly address the immutability trade-off identified in Section 5.7 (Table 5), which shows that immutability, often treated as an unconditional benefit, can entrench errors precisely because it forecloses correction.

6.5. Limitations and Future Research

This study has several methodological limitations. First, although the literature synthesis used defined eligibility criteria, targeted database searching, citation chaining, source verification, and structured thematic coding, complete database-specific search logs and intermediate screening records were not retained. The search process therefore cannot be reproduced at the level expected of a formal PRISMA-based systematic review, and no retrospective reconstruction of unverifiable retrieval, duplicate-removal, or screening counts was attempted; the 28-study corpus should accordingly be interpreted as a structured and targeted synthesis of relevant evidence rather than an exhaustive systematic representation of all literature in the field. Second, thematic coding was conducted by a single coder; although predefined categories and a structured extraction matrix were used to support consistency, the absence of independent duplicate coding and inter-coder reliability assessment may introduce interpretive subjectivity. Third, a formal study-level quality score was not applied because the analytical corpus includes heterogeneous study designs—including systematic and bibliometric reviews, conceptual studies, and application-oriented research—for which a single common appraisal instrument would not provide directly comparable assessments; this limits the ability to weight findings according to a uniform methodological-quality metric. Fourth, the study relies entirely on secondary data and publicly available case material, with no primary empirical investigation of operational AI–blockchain systems in live financial environments. The pace of technological change may also affect how relevant some of the specific applications discussed here remain over time. The geographic coverage skews toward North America, Europe, and East Asia, which may limit how well the findings apply in other regulatory and cultural contexts.

Future research should work on addressing these gaps: longitudinal empirical studies measuring real-world adoption outcomes, deeper investigation of ethical and social implications including algorithmic bias in immutable systems, exploration of how AI–blockchain integration intersects with emerging technologies such as quantum computing and Web3 protocols, and comparative regulatory analysis across a wider range of jurisdictions, including the Global South.

6.6. Conclusions

The convergence of Artificial Intelligence and blockchain represents a genuinely transformative frontier in financial services—one with real potential to improve efficiency, transparency, and inclusion, while also introducing new systemic, ethical, and regulatory challenges that cannot be ignored. This study offers a multi-level theoretical and evidence-informed foundation to help scholars, practitioners, and policymakers navigate that landscape. Realising what AI–blockchain convergence is capable of will take more than technological innovation; it will require organisational readiness, regulatory foresight, and genuine ethical stewardship. By bringing together evidence from the literature and from case study analysis, this paper offers a framework for understanding, implementing, and governing AI–blockchain systems in finance.