The problem facing libraries today is multidimensional. Procedural governance has advanced more rapidly than substantive equity: while transparency mechanisms and auditability protocols are gaining traction, fairness and ethics remain conceptually vague and empirically underexamined. Responsibility is distributed across vendors, IT departments, and library professionals, creating gaps in accountability and in the capacity for ethical redress when algorithmic systems fail or bias emerges. Furthermore, explanation and interpretability often remain inaccessible to practitioners and users, limiting their ability to exercise agency or informed consent in their interactions with automated systems. These challenges reflect deeper tensions between efficiency and justice, automation and stewardship, and innovation and accountability.

A review of the literature reveals significant gaps. Few studies provide empirical evaluations of FATE in practice within library contexts. There is an absence of tailored ethical audit methodologies capable of evaluating bias, provenance, and representational equity in academic data repositories. Fairness metrics are often borrowed from technical AI fields and lack calibration to scholarly or cultural heritage contexts. Explanatory systems, such as human-centred interfaces for cataloguing or recommendation algorithms, are rarely evaluated in terms of user comprehension or trust. Moreover, documentation and provenance practices remain inconsistent, with limited access to training data, model architectures, or change logs, constraining transparency and reproducibility in academic library settings.

This article contributes to the field in four key ways. It offers a systematic mapping of how FATE has been conceptualised and applied within AI-enabled data curation in academic libraries; it identifies the asymmetrical operationalisation of FATE, showing the procedural dominance of accountability and transparency over fairness and ethics; it introduces a conceptual FATE curation framework that embeds ethical checkpoints across data lifecycle stages; and it proposes a research agenda that prioritises domain-specific ethical metrics, participatory audits, and measurable outcomes for responsible AI in librarianship. By doing so, it advances a scholarly foundation for ethical governance that is evidence-based, actionable, and sustainable.

The study focuses specifically on academic and research libraries, analysing contexts where AI influences selection, organisation, preservation, or access to data. It synthesises literature published between 2020 and 2025 that explicitly or implicitly engages with the FATE principles. While policy and regulatory frameworks such as the EU AI Act or UNESCO’s AI Ethics Recommendation provide contextual background, the analysis remains centred on library implementations and evaluative practices. Methodological details, including database selection, screening processes, and PRISMA-based reporting, are presented in the Methodology section.

Two propositions structure this review: first, that accountability and transparency mechanisms are more frequently operationalised than fairness and ethics in library AI deployments; and second, that empirical evidence demonstrating the practical impact of FATE principles remains limited relative to conceptual and policy discourse. These propositions guided the synthesis and evaluation of the reviewed literature. The structure of the article proceeds accordingly: following the methodology, the state of the art explores the conceptual intersections between AI, data curation, and ethics, tracing the evolution of FATE in academic information environments. The results present temporal and thematic trends, leading to a conceptual synthesis in the form of the FATE curation framework. The discussion then interprets the findings considering theoretical and practical implications, concluding with a forward-looking research and policy agenda.

Ultimately, this study argues that in the era of AI-mediated librarianship, governance is inseparable from design. Libraries will uphold epistemic justice and institutional integrity not by appending ethical guidelines after technological adoption, but by embedding fairness, accountability, transparency, and ethics directly into the processes and infrastructures through which knowledge itself is curated.

In academic settings, data curation has shifted from a narrow, repository-centred activity to a strategic, institution-wide function that spans the full research lifecycle and interfaces with algorithmic systems. Early emphases on file-level management and compliance have given way to stewardship models that integrate policy, infrastructure, training, and ethical governance. Recent syntheses in library and information science confirm this widening scope, from supporting researchers’ deposit and documentation practices to shaping the quality, provenance, and responsible reuse of research data within intelligent services (Sheikh et al., 2025; Stvilia et al., 2025; Zareef & Jabeen, 2025). This evolution is inseparable from the embedding of AI in scholarly communication ecosystems, recommendation, discovery, summarisation, and conversational support, thereby relocating curation from a post-hoc preservation task to an upstream, designtime determinant of how knowledge is produced, surfaced, and evaluated (Asemi et al., 2021; Afjal, 2023; Buitrago-Ciro et al., 2025).

Research data management (RDM) concerns policies, planning, documentation, storage, sharing, and compliance at project level; it is primarily operational and researcher-facing (Sheikh et al., 2025). Digital preservation ensures long-term accessibility, authenticity, and integrity through format strategies, fixity, and succession planning; it is archival and time-oriented (Stvilia et al., 2025). Data curation sits across and beyond both: it is the continuous, value-adding process that selects, normalises, enriches, validates, documents, and exposes data for discovery and reuse, aligning technical processes with scholarly meaning and policy goals. In libraries, curation is best understood as a lifecycle that coordinates selection, ingest, description, quality assurance, preservation, and access, often formalised through curation lifecycle models operationalised in library contexts (Stvilia et al., 2025; Zareef & Jabeen, 2025). Clarifying these distinctions matters: research data management can be compliant yet shallow; preservation can be durable yet opaque; curation is the connective tissue that renders data fit for purpose in human and algorithmic workflows.

The lifecycle perspective remains central to understanding data curation in academic libraries. Among the most widely adopted frameworks, the Digital Curation Centre (DCC) model (Figure 2) articulates the iterative processes through which data are conceptualised, created, appraised, curated, preserved, and reused. It emphasises that curation is not a terminal activity but a continuous stewardship function that integrates preservation planning, community participation, and access governance across time. While the Digital Curation Centre has played a foundational role in structuring curatorial thinking within libraries, it remains largely epistemically and ethically neutral. As such, it offers limited guidance for addressing algorithmic bias, accountability, or transparency in AI-mediated data environments.

This model underpins much of the contemporary discourse on data stewardship. It provides the structural foundation upon which newer frameworks, such as FAIR principles and FATE, build to integrate ethical, algorithmic, and accountability dimensions into curatorial practice.

Academic libraries increasingly function as institutional mediators that translate between the practices of researchers, the constraints of platforms, and the expectations of publics. They convene policy (e.g., governance, licensing), build capability (data and AI literacy), and align infrastructure (repositories, knowledge graphs, APIs) with disciplinary norms. Evidence from studies of higher education libraries and organisational adoption work highlight how libraries broker cloud and AI technologies while maintaining service reliability and compliance (Ibrahim et al., 2025). On the human-factors side, libraries are re-positioning staff as data stewards and AI-literate professionals, with emerging guidance on staff competencies and patron education (Ali & Richardson, 2025; Li et al., 2025; Tang, 2025). In short, libraries are no longer only endpoints for data; they shape the conditions of possibility for responsible data-intensive scholarship.

Because curated data feed algorithmic systems, curation choices are epistemic, they structure what can be known, and ethical, they distribute risks and benefits. The FATE principles render this dual nature explicit. Fairness requires representational adequacy and bias mitigation in selection and description; accountability demands governance, audit trails, and answerability for curatorial and algorithmic decisions; transparency calls for documentation that is comprehensible to both experts and affected users; and ethics frames proportionality, consent, privacy, and legitimate interests across the lifecycle. Contemporary reviews in AI ethics and AI auditing show that operationalising these principles hinges on concrete artefacts (documentation, metrics, oversight processes) rather than declarations (Laine et al., 2024; Matei et al., 2025; Memarian & Doleck, 2023).

Within library and information science (LIS), the proliferation of AI-enabled discovery and conversational systems amplifies these stakes: system behaviour can be persuasive, error-prone, or strategically opaque, which raises questions of truthfulness, explainability, and responsible deployment in scholarly mediation (Black, 2024; Buetow & Lovatt, 2024; Kousha & Thelwall, 2024; Matsieli & Mutula, 2025; Njiru et al., 2025; Ridley, 2025; Sparkman & Witt, 2025). Positioning curation as epistemology-in-practice aligns library work with FATE not as an add-on but as a design constraint for data pipelines and intelligent services.

Taken together, these strands justify treating data curation in academic libraries as a socio-technical, lifecycle-based stewardship that is analytically separable from research data management and digital preservation, while remaining interdependent with both. As AI systems mediate scholarly discovery and evaluation, curation becomes the primary locus where FATE is made operational, through selection policies, metadata schemas, documentation standards, quality controls, and governance structures that render data not only findable and durable but just, answerable, and intelligible in algorithmic contexts.

The FAIR principles have guided much of the discourse around data stewardship in academic libraries. However, their conceptualisation often remains technical rather than ethical. The visual metaphor presented below encapsulates the layered and processual nature of FAIR-compliant data management, serving as a conceptual bridge toward FATE-oriented approaches.

Although the FAIR principles provide a robust technical framework for data interoperability and reuse, they do not explicitly address questions of power, bias, responsibility, or ethical governance. This limitation becomes particularly salient in AI-driven contexts, where automated decision-making systems actively mediate access to knowledge, thereby necessitating an explicit ethical lens such as FATE.

The conceptual trajectory traced above can be visually synthesised through a temporal model that illustrates the progressive integration of ethical and algorithmic stewardship into the curatorial mission of academic libraries. Figure 4 (below) captures this evolution from Data Management to AI-Enabled Stewardship between 2020 and 2025, highlighting how the FATE principles have successively permeated the core functions of data practice.

This visualisation reinforces the argument that data curation has not evolved in isolation but in tandem with the growing ethical sophistication of academic information systems. The transition from operational management to value-driven stewardship underscores that contemporary curation operates simultaneously as a technical, epistemic, and moral enterprise: the site where institutional accountability meets algorithmic design.

Over the last five years, academic libraries have moved from experimental pilots to the systematic embedding of AI across cataloguing, subject classification, discovery, and recommendation. Intelligent library initiatives integrate rule-based and machine-learning approaches for metadata enrichment, entity recognition, and automated classification (Asemi et al., 2021). Discovery layers and recommender services increasingly rely on behavioural and semantic models to rank and surface resources, while conversational agents, from general large language models (LLM) to domain-tuned chatbots, mediate reference services, search refinement, and instruction at scale (Aboelmaged et al., 2025; Afjal, 2023; Khosrawi-Rad et al., 2023). In parallel, AI augments the scholarly communication pipeline, screening, peer review support, summarisation, and integrity checks, reconfiguring the informational environment in which libraries operate (Buitrago-Ciro et al., 2025; Carabantes et al., 2023; Kousha & Thelwall, 2024;). These trajectories are consistent with broader information science and user experience evidence: AI is becoming infrastructural, shifting from peripheral add-ons to core components of information services (Collins et al., 2021; Li et al., 2025; Stige et al., 2024).

Automation unsettles the library’s traditional stance of impartial mediation. Algorithmic ranking and classification encode value judgements, about relevance, authority, and visibility, that were once explicit, negotiated, and attributable to human cataloguers. As services become AI-by-default, the locus of decision-making moves upstream into data pipelines and models, potentially obscuring the deliberative practices that sustained professional neutrality (Ibrahim et al., 2025; Mitha & Omarsaib, 2025). Evidence from library and higher education contexts shows that adoption drivers (efficiency, scale, responsiveness) collide with the need for professional oversight, contestability, and pedagogical intent, demanding new literacies and role redesign rather than simple tool substitution (Ali & Richardson, 2025; Romero & Chagua, 2025; Thong et al., 2025). In short, mediation is not disappearing; it is being reconstituted, from front-stage interaction to the design and governance of algorithmic systems.

As curated data feed the very systems that organise scholarly knowledge, curation practices now bear epistemic consequences: what is selected, described, and linked shapes what becomes discoverable and credible. Reviews on human-centred explainability and data-quality assurance emphasise that trust is not a property of models alone but of the end-to-end socio-technical assemblage, documentation, provenance, validation, and communicability to stakeholders (Ridley, 2025; Stvilia et al., 2025). In reference and instructional contexts, LLM-mediated interactions can be useful yet performatively confident, pressuring librarians to establish verifiable pathways, boundaries of use, and escalation protocols (Buetow & Lovatt, 2024; Sparkman & Witt, 2025). Ethical reflections from the library field further argue that the credibility of library services in an AI-saturated ecosystem depends on explicit norms for responsible deployment and on cultivating AI literacy among staff and patrons (Ali & Richardson, 2025; MatsieLi & Mutula, 2025; Tang, 2025).

The expansion of AI amplifies a familiar risk set. Opacity arises from proprietary models, complex pipelines, and the entanglement of third-party services, limiting auditability and user comprehension (Laine et al., 2024). Bias emerges through historical collections, skewed training data, and feedback loops in usage logs, with downstream effects on the visibility of languages, topics, and communities (Memarian & Doleck, 2023; Ofosu-Ampong, 2024). Explainability remains uneven: local, human-centred explanations are often absent or unusable for library stakeholders (Ridley, 2025), while accountability diffuses across vendors, platforms, and institutions, complicating redress when harms occur (Ashok et al., 2022; Heyder et al., 2023). Critical scholarship has even questioned truthfulness and strategic deception in AI outputs, underscoring the need for verifiable evidence paths in scholarly mediation (Black, 2024; Njiru et al., 2025). Together, these findings reinforce the necessity of FATE-aligned governance: bias assessment and representational checks (Fairness), clear responsibility and escalation chains (Accountability), actionable documentation and model reporting (Transparency), and proportionate, context-sensitive deployment grounded in professional ethics (Ethics).

To elucidate this convergence between functionality and ethical oversight, Table 1 maps the main domains of AI application within academic libraries and aligns them with their corresponding FATE-related risks and mitigation mechanisms. The comparative structure exposes the dual nature of algorithmic integration: while it augments efficiency, personalisation, and scalability, it simultaneously generates new vectors of opacity, bias, and ethical vulnerability.

Taken together, the analysis suggests that ethical risk intensifies in proportion to the epistemic mediation exercised by algorithmic systems. Applications such as discovery algorithms and conversational agents emerge as high-risk zones, as they directly shape information visibility, relevance, and user interpretation while often operating under conditions of limited transparency.

To synthesise these dynamics, Table 1 provides a comparative overview of the main AI functions currently adopted in library operations, mapping their primary benefits against associated ethical risks and corresponding governance mechanisms. This structured view clarifies how each AI-driven functionality simultaneously advances service performance and introduces vulnerabilities that require deliberate institutional oversight and human–algorithmic co-accountability.

Within LIS, the ethical stakes of data curation are no longer peripheral. As curated data now drive intelligent services, curation choices have epistemic and moral consequences, shaping what can be known, by whom, and under which conditions. Recent syntheses emphasise that credible, human-centred information services require explicit alignment between design, documentation, and stakeholder needs (Li et al., 2025; Stvilia et al., 2025). Cross-domain reviews show that FATE is the most practical lens for translating broad AI ethics into operational requirements for higher-education and library contexts (Ashok et al., 2022; Memarian & Doleck, 2023). In parallel, conceptual work on university data culture argues that institutions must articulate values and norms that travel consistently from policy to pipeline to interface (Tang, 2025).

AI adoption promises efficiency, scale, speed, and responsiveness across classification, discovery, and service triage (Asemi et al., 2021; Collins et al., 2021), but the literature repeatedly documents tensions with public-interest values central to academic libraries. Recommenders and LLM-based agents can entrench representational imbalances, privileging dominant languages, topics, or venues, thereby undermining equity and inclusion in scholarly visibility (Memarian & Doleck, 2023; Ofosu-Ampong, 2024). Accessibility is similarly at stake: when interfaces are optimised for engagement rather than intelligibility, users with different literacies or access needs may be disadvantaged (Stige et al., 2024). The emerging response within the sector foregrounds AI literacy, for staff and patrons, as a precondition for responsible deployment and for sustaining trust in mediated services (Ali & Richardson, 2025; Mitha & Omarsaib, 2025).

Converging strands of evidence point to governance, not tooling, as the decisive variable. Reviews on ethics-based AI auditing propose translating principles into auditable artefacts: risk registers, model cards, datasheets, issue-tracking, and escalation protocols mapped to accountable roles (Laine et al., 2024). Broader information-management frameworks stress integrating ethics by design into procurement, vendor management, and lifecycle controls (Ashok et al., 2022). Within LIS, studies highlight the importance of quality assurance, provenance, and documentation in repositories and services as the substrate for any trustworthy AI layer (Stvilia et al., 2025; Zareef & Jabeen, 2025). Organisational adoption research in university libraries further shows that cloud/AI decisions hinge on institutional capability, governance maturity, and human-factor readiness, underscoring the need for clear responsibility chains and competence development (Ibrahim et al., 2025; Ramirez et al., 2022).

The record is mixed. Conversational agents improve reach and responsiveness, yet raise concerns about privacy, misdirection, and boundary-of-competence signalling (Aboelmaged et al., 2025). Generative tools assist summarisation and literature triage, but invite over-reliance, untraceable claims, and goal-shift from evidence to fluency (Afjal, 2023; Buetow & Lovatt, 2024; Sparkman & Witt, 2025). In scholarly communication, AI-enabled peer-review support and screening can streamline workflows while complicating accountability and authorship norms (Carabantes et al., 2023; Kousha & Thelwall, 2024). Studies of user profiling warn of opaque inferences and disproportionate data capture with unclear redress, directly engaging FATE fairness and accountability planks (Njiru et al., 2025). Adoption of robotic and automated service components surfaces labour, safety, and legitimacy questions that libraries are not yet institutionally equipped to govern at scale (Shahzad et al., 2024). At the conceptual core lie opacity, bias, explainability, and diffused responsibility: human-centred XAI (explainable artificial intelligence) remains unevenly available in library-facing tools (Ridley, 2025), and philosophical critiques underscore that systems can produce persuasive falsehoods with high confidence, demanding verifiable evidence paths in scholarly mediation (Black, 2024; Buitrago-Ciro et al., 2025).

Taken together, these strands motivate a governance posture in which fairness is operationalised via representational audits and impact assessment; accountability via role-bound decision logs, vendor contracts, and incident response; transparency via public-facing documentation, data provenance, and explanation mechanisms suitable for library stakeholders; and ethics via proportionality, consent, privacy-by-design, and professional standards embedded across the curation lifecycle. In AI-driven data curation, governance is not ancillary to service quality; it is the mechanism by which libraries reconcile efficiency with public interest values, sustaining informational trust in an algorithmic environment.

To consolidate these findings and translate them into an actionable governance structure, Figure 5 illustrates the sequential embedding of the FATE principles across institutional workflows in AI-driven data curation. It depicts how FATE can be operationalised at each decision point, from policy formulation to data ingestion, metadata enrichment, and access management, thereby transforming governance from a reactive control mechanism into an active design principle.

Rather than serving as a descriptive overview of existing practices, the following figure should be read as a prescriptive governance scaffold. It delineates critical intervention points at which FATE principles can be institutionally embedded across the AI-enabled data curation workflow.

This model provides a conceptual bridge to the next section, which examines in greater depth the theoretical origins, interpretations, and operational applications of the FATE framework within academic librarianship.

FATE emerged as a pragmatic distillation within AI ethics to bridge abstract normative claims and implementable requirements in socio-technical systems. In information systems and human-AI interaction research, this cluster functions as a practice-oriented vocabulary capable of travelling across disciplines and stakeholder groups (Collins et al., 2021; Heyder et al., 2023). Higher education reviews explicitly codify FATE as the dominant lens for aligning AI deployments with institutional duties of care (Memarian & Doleck, 2023), while management scholarship frames FATE as the anchor for governance interventions, risk registers, role assignments, documentation regimes, and audit trails, that render ethical claims operational (Ashok et al., 2022; Laine et al., 2024). Within LIS, this translation is amplified by the fact that curated datasets feed intelligent services: curation choices thus carry epistemic and moral weight in how knowledge is produced, surfaced, and evaluated (Li et al., 2025; Stvilia et al., 2025).

While FATE provides an actionable ethics lens for AI systems, academic libraries operate within broader data justice traditions that emphasise community rights and collective benefit. The CARE principles (collective benefit, authority to control, responsibility, and ethics), emerged from Indigenous data sovereignty and extend beyond procedural compliance to relational accountability, equitable value sharing, and culturally grounded consent. Positioning CARE alongside FATE reframes data curation as a socio-technical practice: not only how data are made findable and reusable, but for whom, under whose authority, and to what ends. This perspective is especially salient for libraries stewarding sensitive, community-originated, or equity-relevant datasets.

By juxtaposing CARE with FATE, the study foregrounds the need for hybrid governance models that fuse technical accountability with community-centred ethics. In the sections that follow, FATE functions as the operational scaffold for AI-mediated curation, while CARE informs the distribution of authority, consent practices, and benefit allocation, together shaping a coherent approach to responsible data stewardship in academic libraries.

Building upon the preceding discussion of governance mechanisms, the following table operationalises the FATE framework within the specific context of academic libraries. It maps each principle to its functional definition, practical instantiation, and representative scholarly sources. By providing this structured overview, Table 2 transforms FATE from a normative construct into an auditable schema of practice, offering a replicable reference model for institutional adoption and evaluation.

Importantly, the FATE principles should not be interpreted as an isolated checklist. The literature consistently indicates their interdependence, whereby weaknesses in transparency, for instance, undermine accountability, and failures of fairness often signal deeper ethical and governance deficiencies.

Although not the focus of our corpus, widely cited policy instruments, OECD AI Principles, UNESCO Recommendation on the Ethics of AI, the EU AI Act’s risk-based approach, and IEEE Ethically Aligned Design, map naturally onto FATE: non-discrimination and social equity (fairness), institutional governance and redress (accountability), documentation and intelligibility (transparency), and human-rights-grounded proportionality (ethics). Ethics-based auditing literature makes this bridge explicit by translating high-level norms into auditable artefacts (model cards, datasheets, issue logs, escalation pathways) and stakeholder responsibilities (Laine et al., 2024). In academic libraries, these mappings are mediated by sectoral policies (procurement, vendor management, privacy) and capacity-building agendas (AI literacy), which determine the enforceability of FATE beyond aspiration (Ali & Richardson, 2025; Tang, 2025).

Fairness remains contested: definitions oscillate between distributive goals (who is represented and how) and statistical criteria in model behaviour; both are sensitive to data provenance and feedback loops in usage logs (Memarian & Doleck, 2023; Ofosu-Ampong, 2024). Accountability is frequently diffused across vendors, platforms, and institutions, risking responsibility gaps unless decision rights, auditability, and redress mechanisms are contractually and procedurally specified (Ashok et al., 2022; Heyder et al., 2023). Transparency is multi-layered, data, model, and process, and often collapses into disclosure without comprehension; human-centred explainability is necessary if documentation is to be actionable for library stakeholders (Ridley, 2025; Stvilia et al., 2025). Ethics can drift into a residual category or a signal of virtue unless anchored in proportionality, privacy-by-design, consent practices, and clear boundaries of use (Laine et al., 2024; Matei et al., 2025). Critical accounts further warn that generative systems may produce confident falsehoods or strategically misleading content, sharpening the stakes for verifiability and duty of care in scholarly mediation (Black, 2024; Buetow & Lovatt, 2024; Sparkman & Witt, 2025).

For data-intensive scholarship, FATE provides an actionable scaffold across the curation lifecycle. Fairness entails representational audits and bias checks in selection and description, especially across languages, disciplines, and communities; accountability requires role-bound decision logs, procurement clauses, incident response, and oversight boards; transparency calls for provenance-rich metadata, public-facing documentation (e.g., datasheets, model cards), and explanation mechanisms commensurate with user needs; ethics mandates proportional data capture, privacy safeguards, consent management, and clear limits on automation in instructional and reference contexts (Ali & Richardson, 2025; Sheikh et al., 2025; Stvilia et al., 2025). Sectoral evidence shows that these controls are most effective when embedded as infrastructure rather than add-ons, i.e., when governance, training, and platform choices are codesigned with curatorial workflows (Ibrahim et al., 2025; Zareef & Jabeen, 2025). In practice, conversational systems, profiling services, and automated discovery expose libraries to concrete FATE dilemmas, privacy and bias risks in user modelling, opacity in ranking, and explainability gaps at the point of need, requiring measurable safeguards and escalation pathways (Aboelmaged et al., 2025; Buitrago-Ciro et al., 2025; Kousha & Thelwall, 2024; Njiru et al., 2025).

As a boundary object, FATE is rigorous enough to guide design and audit, yet flexible enough to align librarians, vendors, administrators, and researchers around shared commitments. Its value for academic libraries lies in turning ethical intent into curatorial practice, policies, documentation, metrics, and accountability structures that make data not only durable and discoverable but also just, answerable, and intelligible in algorithmic environments.

In academic libraries, FATE can be rendered operational by aligning each curatorial stage with explicit safeguards and artefacts. In selection and acquisition, Fairness entails representational checks across languages, disciplines and communities; Accountability requires decision logs and role-bound sign-off; Transparency is served by public selection policies and provenance notes; Ethics demands proportionality and rights assessment for sensitive data (Sheikh et al., 2025; Stvilia et al., 2025). In ingest and normalisation, libraries implement quality gates, record lineage, and define escalation routes for exceptions, turning pipeline control into auditable practice (Stvilia et al., 2025). In description and enrichment, bias-aware subject vocabularies and authority control address representational harms; documentation of transformations sustains explainability for downstream AI (Li et al., 2025; Zareef & Jabeen, 2025). For access, discovery and recommendation, libraries can conduct exposure audits on ranking outcomes, monitor error profiles for subgroups, and provide user-facing rationales or boundaries of competence for conversational agents and recommenders (Aboelmaged et al., 2025; Ridley, 2025). Organisationally, AI literacy programmes and procurement clauses translate ethical intent into practice by conditioning deployment on documentation, monitoring, and redress (Ashok et al., 2022; Ali & Richardson, 2025).

Three friction points recur. First, vendor opacity: proprietary models and entangled third-party services limit access to training data, features, and evaluation protocols, constraining both transparency and independent audit (Laine et al., 2024). Second, diffused responsibility: accountability is split across libraries, IT units, and suppliers; without explicit decision rights, incident response, and dispute resolution, accountability collapses into accountability theatre (Heyder et al., 2023; Ibrahim et al., 2025). Third, metric misfit: statistical fairness criteria travel poorly into cultural-heritage and scholarly discovery contexts; libraries need domain-appropriate indicators (exposure diversity, language coverage, authority balance) rather than generic parity scores (Memarian & Doleck, 2023; Ofosu-Ampong, 2024). These constraints are amplified by workload and capability pressures, where efficiency incentives risk crowding out deliberation on inclusion, accessibility, and proportionality (Mitha & Omarsaib, 2025; Stige et al., 2024).

Ethics-based auditing proposes a concrete toolkit that libraries can adopt and adapt: model cards for discovery and recommendation services (objective, inputs, known limitations, appropriate/forbidden uses), datasheets for datasets across metadata pipelines (origin, coverage, transformations, quality tests), and algorithmic audits combining offline tests (synthetic queries, counterfactual prompts) with online monitoring (exposure diversity, complaint analytics, drift) (Laine et al., 2024; Stvilia et al., 2025). For conversational systems and LLM-based services, boundary-of-competence statements, verifiable citation pathways, and escalation protocols operationalise transparency and accountability at the interface (Aboelmaged et al., 2025; Afjal, 2023; Sparkman & Witt, 2025). At the institution level, risk registers, role matrices, change logs, and post-incident reviews embed FATE into routine governance, while capability building (AI literacy for staff and patrons) sustains informed oversight (Ali & Richardson, 2025; Ashok et al., 2022; Tang, 2025).

Despite momentum, several lacunae endure. There is no widely adopted, library-specific FATE metric suite that captures the sociotechnical nature of discovery (e.g., how ranking interacts with disciplinary canons and language hierarchies); current practice relies on ad-hoc indicators with limited comparability (Stvilia et al., 2025; Zareef & Jabeen, 2025). Documentation debt persists libraries inherit tools without sufficient model and/or dataset artefacts to meet stakeholder-oriented transparency (Laine et al., 2024; Ridley, 2025). Governance maturity is uneven, with procurement, vendor management, and policy rarely co-designed with curatorial workflows (Ibrahim et al., 2025). Finally, ethical intent often outruns operational capacity: staff time, skills and authority to contest vendor defaults remain constrained, especially where LLM-based features are marketed as turnkey solutions (Buitrago-Ciro et al., 2025; Kousha & Thelwall, 2024).

Conceptual work in LIS and adjacent fields argues for treating FATE as infrastructure, not addon, embedding ethical controls into schemas, workflows, and platforms, and auditing them as part of routine quality assurance. Empirical and policy-oriented studies further recommend coproduction with vendors and researchers to align evaluation datasets, impact metrics, and explanation modalities with scholarly use cases (Laine et al., 2024). In practice, this means, mapping FATE to the curation lifecycle; specifying minimum documentation artefacts per stage; adopting exposure and representational indicators alongside technical metrics; and establishing clear responsibility chains with escalation thresholds and redress routes. Taken together, these steps convert FATE from aspiration into a workable governance grammar for AI-mediated data curation in academic libraries.

Moving beyond theoretical articulation, the following model integrates the FATE framework into the data curation lifecycle, transforming abstract principles into operational checkpoints. Figure 7 illustrates this integration as a cyclical ecosystem where ethical reasoning and technical execution co-evolve. By embedding fairness, accountability, transparency, and ethics into each stage, from dataset selection to end-user access, the model reframes curation as a dynamic process of continuous moral calibration rather than a static technical routine.

Figure 6 represents an original conceptual synthesis derived from the findings of the systematic literature review, integrating curatorial stages with FATE principles and associated operational artefacts.

This systemic alignment not only strengthens institutional integrity but also redefines the professional identity of academic librarians as algorithmic stewards, responsible for ensuring the epistemic and ethical legitimacy of curated knowledge infrastructures.

To capture the evolving maturity of ethical governance across AI-enabled library ecosystems, Table 3 synthesises three prevailing models of institutional response: reactive compliance, proactive stewardship, and embedded ethics. These approaches represent a progressive shift from procedural conformity to value-oriented accountability. The comparison illuminates the extent to which libraries internalise FATE principles not merely as abstract ideals but as operational norms. This transition from compliance to embedded ethics constitutes a pivotal transformation in the governance of AI-mediated data environments and provides a conceptual bridge to the subsequent mapping of FATE-curation intersections.

The increasing interdependence between AI, data curation, and ethical governance has created a dense and evolving conceptual landscape. To elucidate this complexity, Figure 7 presents a conceptual map that visualises the convergence of AI functions, curatorial processes, and the FATE principles. The visual synthesis highlights both the emerging synergies and the persistent gaps, particularly the underrepresentation of fairness and ethics within current library and information science literature.

The systematic review reveals a rapidly expanding but uneven research landscape at the intersection of data curation, AI and ethical governance between 2020 and 2025. Publication activity peaks in 2023-2025, reflecting the surge of generative AI and its implications for scholarly communication, information infrastructures, and professional accountability (Aboelmaged et al., 2025; Buetow & Lovatt, 2024; Buitrago-Ciro et al., 2025). This temporal distribution suggests a paradigm shift: from discussions of automation and efficiency (Asemi et al., 2021; Collins et al., 2021) toward ethical operationalisation and socio-technical responsibility in academic library contexts (Ali & Richardson, 2025; Zareef & Jabeen, 2025).

Across the corpus, accountability and transparency emerge as the most consolidated principles within the FATE framework. Studies on AI auditing and governance (Ashok et al., 2022; Laine et al., 2024) emphasise traceability, model explainability, and stakeholder disclosure as the main ethical practices currently achievable in applied settings. The notion of traceable intelligence (Ridley, 2025) and the creation of documentation artefacts such as datasheets, algorithmic logs, and provenance chains reflect an empirical bias toward procedural transparency rather than substantive fairness. In parallel, library-focused research reinforces the librarian’s evolving role as an ethical intermediary who curates not only datasets but also the metadata and contextual knowledge that enable algorithmic accountability (Sheikh et al., 2025; Stvilia et al., 2025).

While fairness and ethics appear in conceptual and normative discussions (Memarian & Doleck, 2023; Matei et al., 2025), empirical integration into library systems and data workflows remains scarce. Ethical reasoning tends to be externalised, delegated to institutional policies or compliance frameworks, rather than embedded within the mechanics of data stewardship. The literature reveals few attempts to quantify or operationalise equity within metadata schemas, discovery algorithms, or access protocols, highlighting a methodological blind spot between declarative ethics and measurable outcomes (Matsieli & Mutula, 2025; Njiru et al., 2025).

A notable conceptual synthesis emerges around terms such as algorithmic stewardship, ethical metadata and responsible curation. These constructs reflect a move beyond abstract ethics toward embedded governance, where librarianship intersects with AI ethics, information management, and digital policy (Ali & Richardson, 2025; Tang, 2025). Algorithmic stewardship reframes librarians as active custodians of the full AI pipeline, from data acquisition to model validation and interpretability, ensuring that decision-making processes remain transparent, contestable, and auditable (Lee et al., 2023; Roy, 2025). Similarly, ethical metadata practices propose value-sensitive vocabularies and annotations that make datasets readable not only by machines but also by moral and regulatory frameworks (Li et al., 2025).

To consolidate these patterns, a FATE-curation matrix is proposed, aligning stages of the data curation lifecycle (selection, documentation, preservation, dissemination) with the four FATE principles. Such a conceptual map exposes the areas of highest maturity, chiefly transparency and accountability in repository management, and the critical gaps in fairness and ethics integration. This visual and analytical synthesis underscores that while the field advances toward procedural accountability, epistemic justice and ethical embedding remain the next frontier for research and practice in AI-driven academic librarianship.

The synthesis of the reviewed literature exposes a persistent asymmetry between the normative articulation of FATE principles and their practical implementation within academic library ecosystems. While ethical aspirations are increasingly present in strategic frameworks and theoretical discussions (Ashok et al., 2022; Laine et al., 2024; Memarian & Doleck, 2023), empirical validation remains remarkably limited. Few studies have examined how fairness, accountability, transparency, and ethics operate as measurable or auditable constructs in library-managed AI systems (Mitha & Omarsaib, 2025; Sheikh et al., 2025).

Most contributions to date remain conceptual, reflective, or policy-oriented, offering prescriptive statements rather than evidence-based demonstrations (Buitrago-Ciro et al., 2025; Ridley, 2025). Experimental and field-based investigations, such as algorithmic audits of discovery tools, recommender systems, or digital repositories, are notably scarce. Even within the growing corpus on data curation, ethical compliance tends to be reported as a descriptive attribute rather than a processual outcome (Stvilia et al., 2025). This gap suggests the need for interventional research designs capable of evaluating how FATE-aligned mechanisms affect access, inclusion, and user trust in library information systems.

A central methodological deficit lies in the lack of AI auditing models tailored to the library and information science domain. Existing frameworks, such as those discussed by Li et al. (2025) and Laine et al. (2024), are often derived from industrial or technical contexts, insufficiently sensitive to the epistemic and social dimensions of information stewardship. No standardized metrics currently exist to assess ethical performance in library AI systems, particularly regarding fairness in metadata representation, accountability in data provenance, or transparency in automated classification. Developing domain-specific auditing tools thus represents an urgent avenue for scholarly and institutional innovation.

The literature calls for integrative architectures that align technological infrastructures with ethical governance mechanisms (Matei et al., 2025; Njiru et al., 2025). Hybrid frameworks, combining algorithmic auditing, policy regulation, and librarian-mediated oversight, are positioned as the next evolution in responsible information management. Such models would operationalise FATE through continuous feedback loops between system design, ethical evaluation, and institutional accountability (Collins et al., 2021; Tang, 2025).

A promising trajectory lies in the consolidation of ethical AI librarianship, a paradigm where librarians act as mediators of algorithmic accountability infrastructures (Ali & Richardson, 2025; Matsieli & Mutula, 2025). This perspective redefines librarianship as an ethical-technical profession grounded in algorithmic literacy, data ethics, and participatory governance. In this vision, academic libraries evolve into living laboratories of responsible AI practice, testing, documenting, and modelling ethical compliance for broader institutional ecosystems.

In summary, future research must transcend descriptive analyses and advance toward empirical, evaluative, and co-creative methodologies that transform FATE from a conceptual ideal into an operational reality. This progression will not only enhance the ethical resilience of academic libraries but also position them as critical actors in the governance of trustworthy, transparent, and socially responsible AI.

To complement the conceptual overview, Figure 8 positions each of the FATE principles within a maturity, validation matrix. This visualisation clarifies the uneven evolution of research maturity across dimensions of fairness, accountability, transparency, and ethics, revealing a dominance of theoretical exploration over empirical validation. The matrix thus provides a diagnostic lens for identifying areas requiring future research investment and methodological development.

The visual imbalance highlighted in the matrix underscores a critical disconnect between conceptual maturity and empirical validation. While accountability and transparency benefit from clearer operational pathways, fairness and ethics remain largely aspirational, signalling an urgent need for applied studies and evaluative metrics within academic library settings.

This section proposes a FATE curation framework that integrates the research data curation lifecycle with operational controls aligned to FATE. The aim is not to add an ethical layer after the fact, but to embed measurable safeguards into the design, documentation, and governance of library data workflows and AI-enabled services (Ashok et al., 2022; Laine et al., 2024).

We treat the curation lifecycle, selection, ingest and normalisation, description and enrichment, preservation, and access and discovery, as the locus where FATE can be made actionable. Each stage is associated with concrete artefacts, decision rights, and evaluation routines that render ethical commitments auditable and explainable to stakeholders (Li et al., 2025; Stvilia et al., 2025):

Four stages require heightened attention due to their downstream impact and risk concentration:

The framework is immediately applicable where libraries control repositories, metadata pipelines, and procurement terms, contexts in which governance artefacts (risk registers, change logs, incident reviews) can be mandated alongside technical controls (Ashok et al., 2022; Laine et al., 2024). Limitations arise with vendor opacity and entangled third-party services, which constrain auditability and explanation (Lee et al., 2023). Organisational capacity, skills, time, decision rights, also conditions feasibility, underscoring the role of AI literacy and data culture in sustaining implementation (Ali & Richardson, 2025; Ibrahim et al., 2025; Tang, 2025). Finally, metric transferability is non-trivial: generic fairness statistics often misfit scholarly discovery and cultural-heritage settings, calling for domain-calibrated indicators (e.g., exposure diversity, language coverage, provenance completeness) (Ofosu-Ampong, 2024; Stige et al., 2024).

Moving from aspiration to evidence requires interventional and evaluative designs:

This synthesis positions FATE as infrastructure for data curation: a set of embedded controls, artefacts, and roles that make AI-enabled library services not only efficient but answerable and intelligible. The ensuing Discussion can therefore:

By articulating control points, applicability conditions, limitations, and validation methods, the FATE curation framework converts ethical intent into a workable governance grammar for AI-mediated academic librarianship.

Building upon the preceding synthesis, Table 4 summarises the operational components of the proposed FATE curation framework. It aligns curatorial stages with their corresponding FATE principles, key artefacts, and evaluation metrics. This tabular synthesis provides a replicable foundation for both scholarly validation and professional implementation in academic libraries.