Cybersecurity incidents pose a growing threat to the integrity, reliability, and auditability of accounting information systems (AIS), yet little is known about how leadership expertise shapes firms’ voluntary disclosure of such events. We conceptualize voluntary cyber-incident disclosure as an outcome of AIS governance, reflecting how firms communicate disruptions to system integrity and reliability. Drawing on Upper Echelons Theory and Voluntary Disclosure Theory, we examine whether domain-specific expertise of audit committee (AC) members and chief executive officers (CEOs) influences voluntary cyber-incident disclosure. Using 7,411 U.S. firm-year observations from 2017–2022, we document systematic asymmetric effects. IT expertise on audit committees and among CEOs is positively associated with voluntary disclosure, consistent with a system-integrity perspective emphasizing transparency in response to AIS disruptions. In contrast, financial expertise is negatively associated with disclosure, suggesting a capital-market framing that prioritizes litigation risk and proprietary costs. Audit committee external directorships and CEO tenure are also positively associated with disclosure. These effects are consistently concentrated across all AC and CEO attributes in unregulated industries, where managerial discretion over disclosure decisions is greater. Instrumental variable analyses support the robustness of our findings. Overall, leadership expertise shapes how cybersecurity risks are cognitively framed, advancing AIS research on digital risk governance.
This study proposes a Large Language Model (LLM)-assisted framework for parsing near real-time audit evidence from text to cross-verify accounting data. This addresses a core challenge in continuous auditing (CA): analyzing textual data in near real-time. The three-step framework involves 1) Preprocessing text, 2) LLM inference guided by auditor objectives/schemas and prompts, and 3) Validating accounting records against LLM-derived audit evidence. Demonstrated on a real-life Brazilian governmental payroll system, the framework cross-verifies payroll data with human resources information from the Official Gazette. Compared to auditors' existing process, the framework improves audit effectiveness through full population testing with cost efficiency, decreases the cross-verification time by up to 83%, and achieves a high extraction accuracy of 97%. The study demonstrates the advantage of LLMs in facilitating audit quality and contributes by proposing an LLM-assisted framework to facilitate practical CA application. Beyond showcasing the potential of LLMs, this study contributes by structuring their application into a replicable three-step framework that auditors without advanced technical expertise can adopt. By aligning LLM capabilities with auditor-defined objectives, parsing schemas, and rule-based cross-verification, the framework ensures that efficiency gains translate into effectiveness, replicability, and audit quality in real CA environments.
This study examines the information content of topic-specific cybersecurity disclosures following guidance from the U.S. Securities and Exchange Commission (SEC), 2011 , Securities and Exchange Commission (SEC), 2018 , Securities and Exchange Commission (SEC), 2023 , with a focus on how disclosure placement shapes informativeness. I develop a scalable topic-level Disclosure Completeness Index that captures the intensity and distribution of disclosures across eight regulator-identified cybersecurity categories. Using 13,800 firm-years from 2019 to 2024, I document substantial cross-topic heterogeneity and show that greater topic-level disclosure intensity is associated with weaker subsequent performance, consistent with disclosures reflecting underlying cybersecurity risk. Crucially, identical topics have systematically different associations depending on placement: Risk Factor disclosures are more strongly and consistently linked to next-year profitability and valuation than analogous narratives in the Management's Discussion and Analysis section. Topic-level analysis further indicates that disclosures related to operational exposure are the primary factors of these associations and are economically meaningful. In contrast, less frequently disclosed topics—such as the probability and magnitude of future incidents—exhibit limited predictive content. Additional analyses show that these patterns vary across industries with different levels of breach exposure. Overall, this study highlights disclosure placement as an economically meaningful dimension of textual reporting and introduces a reusable framework for evaluating the informational content of topic-specific disclosures.
Prior research has documented a number of positive outcomes attributable to managerial ability, such as higher earnings quality, decreased likelihood of going concern opinions, and shorter audit report lags. In this study, we examine the association between managerial ability and cybersecurity breaches, a growing concern for a wide range of stakeholders. We posit that higher managerial ability translates to an improved set of cybersecurity risk mitigation measures, thereby lowering the likelihood of cybersecurity breaches. Using the Demerjian et al. (2012) measure of managerial ability on a sample of 13,813 firms, results indicate that higher managerial ability is more likely to attenuate cybersecurity breach incidents. Considering the effects of managerial ability on internal and external breaches separately, we further find that higher managerial ability is effective at mitigating risks of both breach types. This study contributes to the literature by providing empirical evidence of an association between managerial ability and cybersecurity breaches, for both internal and external breach types.
Corporate carbon disclosure, including Scope 3 emissions, has increased significantly in recent years and has become important to regulators, investors, and other stakeholders. However, Scope 3 data remain challenging to quantify due to their inherent methodological and operational complexity, leading to significant information gaps and unreliable disclosures. To address these challenges, this study proposes a mobile application framework to improve the collection and management of Scope 3 emissions data. The proposed app allows for the collection of real-time data and incorporates interactive digital tools and gamification features to facilitate the collection of granular, activity-level primary data. By obtaining primary Scope 3 emissions data directly from consumers, employees, and suppliers, the proposed app enhances the quality, credibility, and transparency of sustainability disclosures. Enhanced data availability strengthens firms' compliance with reporting requirements and provides stakeholders with more reliable information, thereby improving decision-making. In addition, the application provides managers with more timely emissions information, supporting operational adjustments and decision-making that enhance efficiency and cost management. This study contributes to the sustainable accounting and to the accounting information systems literature by proposing a digital data framework that supports stakeholder-level data integration and enhances the reliability of sustainability information.
Environmental reporting has assumed growing importance in capital markets, particularly in shaping investors’ decision-making. However, the absence of universally accepted reporting standards and domain-specific assurance gives managers substantial discretion to present their sustainability performance in a deceptively favorable way, which is commonly called greenwashing. This risk is further compounded by investors’ limited access to robust data and specialized analytical expertise, which impairs their ability to detect greenwashing in environmental disclosures. Large Language Models (LLMs) with advanced reasoning capabilities and integrated web search functions offer a novel approach to addressing these challenges. This study proposes a framework for developing LLM-based agentic systems to detect greenwashing risks in environmental disclosures. Since greenwashing is not a binary outcome but a multi-dimensional phenomenon, this study introduces seven indicators to assess different facets of greenwashing risks. These indicators are further integrated into a dashboard tailored to the needs of various users. This study contributes to the greenwashing literature by establishing a comprehensive framework for constructing LLM-based agents to detect greenwashing and proposing a set of quantifiable greenwashing indicators. It also contributes to the accounting profession by providing a tool for real-time monitoring of greenwashing risks.
In the increasingly competitive AI chatbot market, user switching behavior poses a significant challenge for AI businesses to achieve their sustainable development. To investigate why users decide to switch, this study establishes a multidimensional influencing factor model based on the Push-Pull-Mooring (PPM) framework. Specifically, privacy risk and credibility risk are identified as push factors; subjective norm and novelty value are pull factors; transition cost, sunk cost, and inertia are mooring factors. This study employs a multi-method approach to analyze survey data from 261 AI chatbot users. The methods include partial least squares structural equation modeling (PLS-SEM), fuzzy-set qualitative comparative analysis (fsQCA), and necessary condition analysis (NCA). By integrating variable-centered and configuration-based perspectives, this study examines the independent effects, synergistic interactions, and necessary conditions of the influencing factors. This approach makes the results solid, complete, and easier to understand. The study contributes to accounting and information systems (IS) by dissecting the antecedents of user switching behavior in the AI chatbot context, using a theoretically grounded and methodologically innovative research design. It extends the PPM model to human-AI interaction, which enriches the theoretical landscape of accounting user behavior in emerging AI technologies. This study also provides empirical evidence to assist AI enterprises in accurately identifying the drivers of user churn and formulating effective user retention strategies. The findings offer valuable implications for cultivating a healthy development environment within AI enterprises.
This study examines the financial informativeness of ESG report narratives by applying large language models (LLMs) to summarize disclosures and conduct sentiment analysis. Using Gemini 1.5 Flash, we generate summaries for ESG reports of non-financial S&P 500 firms from 2010 to 2024 and derive sentiment scores with FinBERT. We find that sentiment extracted from LLM-generated summaries, not the original full texts, is significantly associated with current and future financial performance and forward-looking market valuation. In contrast, institutional ESG ratings show limited explanatory power. Our additional analysis reveals that governance-related sentiment predicts profitability, while social sentiment is linked to valuation. These results suggest that LLMs can reduce information processing costs and amplify decision-useful signals in ESG narratives. This study contributes to ESG research by showcasing how LLM-based tools enhance the interpretability and relevance of voluntary disclosures for capital market participants.
While machine-readable datasets have advanced capital markets research, much decision-relevant information remains embedded in unstructured, PDF-based financial disclosures, limiting the scalability of archival research. Keyword-based heuristics and traditional statistical and machine-learning approaches struggle with the heterogeneity and complexity of financial reporting formats. This study proposes a scalable AI-based framework for extracting structured data from unstructured financial reports. Leveraging a context-aware prompting strategy with large language models (LLMs), the framework identifies relevant page ranges, isolates targeted disclosure sections, and extracts structured quantitative information. We demonstrate its utility by extracting capitalization thresholds from the “Capital Assets” note disclosures in governmental Annual Comprehensive Financial Reports (ACFRs), achieving 99% accuracy in page-range identification and 94% accuracy at the individual note level. We further validate the framework by extracting Valuation Allowances for Deferred Tax Assets from corporate 10-Ks with 100% success. This study contributes to accounting research by enabling large-sample archival analyses using previously inaccessible disclosures, to the Accounting Information Systems (AIS) literature by demonstrating the application of LLMs to complex extraction tasks, and to practice by improving the efficiency of regulatory oversight and public-sector monitoring.
This study provides early evidence on U.S. public companies’ responses to the SEC’s 2023 rule requiring detailed annual disclosures on cybersecurity risk management and governance. Using textual analysis of 3,440 Item 1C disclosures in 10-K filings from 2024, we investigate the determinants of these newly required disclosure characteristics and assess market reactions. Results show variation in disclosure quality—proxied by length, redundancy, and specificity—primarily driven by firm size, financial performance, auditor quality, cybersecurity and litigation risk exposures, and peer practices, though these factors collectively explain only moderate variance. Past cyber incidents, firm digitalization, material IT weaknesses, and tech-firm status show no influence, suggesting strategic discretion persists even under a mandate. Additional analyses show that Item 1C represents new disclosure content rather than a relocation of existing risk disclosures. To assess market reactions, we utilize event studies, analyze cybersecurity-related discussions in earnings call transcripts, and examine investor attention through filing download activity. The results indicate a minimal response from both investors and analysts to these newly mandated disclosures. These insights hold important implications for policymakers regarding the balance between regulatory burden and informational value, as jurisdictions globally adopt similar cybersecurity disclosure policies.
In today’s rapidly evolving business environment, organizations increasingly develop dynamic capabilities to effectively respond to unprecedented challenges and enhance firm performance. Data-driven decision-making (DDDM) has emerged as a critical strategy, enabling organizations to leverage comprehensive data analysis to guide decision-making and promote evidence-based management. Drawing on the dynamic capabilities view (DCV), this study extends existing research by examining how advancements in information and communication technology (ICT) and business analytics (BA) capabilities support DDDM, thereby facilitating the transformation of big data into improved operational performance. Additionally, we investigate the pivotal role of environmental dynamism in promoting DDDM supported by dynamic capabilities.To validate the proposed model, we conducted a survey among senior managers from medium and large Australian organizations operating at varying levels of dynamic capabilities. The findings demonstrate that dynamic capabilities enhance operational performance through the mediating role of DDDM. Moreover, the study also provides novel insights into the moderating effect of environmental dynamism on the relationship between BA capabilities and DDDM, offering a deeper understanding of how external conditions shape DDDM processes. In contrast to the widespread managerial belief that more analytics always leads to more DDDM and better performance, we find that this is not the case in high environmental dynamism.
In the context of pressure to ensure real-time reporting, accounting professionals must increasingly make judgments about using unfamiliar information sources such as big data. In contrast to prior studies that typically focus on data analytics rather than how big data is judged and used, this study investigates how accounting professionals use big data for business-related decision-making. In our 24 semi-structured interviews, we use a vignette for a consistent context as our accounting professionals make six decisions requiring use of two information sources-accounting information and social media as a form of big data. By applying valence theory to quantify the data, findings show how their propositional judgments positively value accounting information and negatively value most characteristics of this big data. Next, when invoking a practical judgment about whether and how to use big data for decision-making, they apply a heuristic, and weigh its 'relevant-value-to-context' (RVC) by anchoring their judgment in accounting information. In correlating findings with their expressed trust and preference to rely on experience when decision-making, this practical judgment appears to be informed by experience. As both judgments value accounting information, this expressed trust in big data appears related to participants' willingness to accept risk rather than assurance about its trustworthiness. Theoretically, findings suggest need to revisit the role of experience and trust in the extended valence framework. Further, whereas practitioner resources focus on how big data's management value chain affects accountants' roles and skills, findings show the importance of existing capabilities, particularly their training and experience.
This research bridges information systems (IS) and accounting by examining how big data, an IS innovation, affects holistic firm performance as defined in accounting and finance. Theories including the resource-based view, knowledge-based view, and dynamic capabilities framework support the idea that big data, as a strategic knowledge resource, enhances the firm’s competitive advantage and responsiveness to rapidly changing environments.Since firms do not publicly disclose big data expenditures, prior studies often resort to small-sample surveys and narrowly defined firm categories, resulting in limited generalizability. To overcome this limitation, we develop novel proxies for big data emphasis by leveraging text analytics techniques, including cosine similarity and BERT-based embeddings, to measure the emphasis on big data within managerial narratives. The attention-based view suggests that such emphasis serves as a proxy for managerial attention and priorities. This proxy yields a large sample of industrially and chronologically diverse observations, extending the generalizability of our results beyond previous studies.We empirically examine the relationship between big data emphasis and market-based (Tobin’s Q) and accounting-based (Return on Assets) holistic performance measures. Updating the literature, we find that management’s emphasis on big data (1) strengthens Tobin’s Q, revealing the market’s optimism about big data, but (2) undermines Return on Assets, reflecting substantial costs of big data.The study’s novelty lies in its objective, scalable measurement of big data emphasis from textual disclosures, overcoming the subjectivity of prior approaches. By highlighting the divergence between market-based and accounting-based outcomes, we provide stakeholders with a balanced framework for evaluating technology adoption.
This research aims to explore the interplay between stakeholders and the US accounting standard setter on the accounting treatment of crypto assets. We applied the beliefs-actions-outcomes (BAO) model to capture stakeholders' expectations, expressed in their comment letters in response to the Financial Accounting Standards Board's Exposure Draft (FASB's ED) on Topic 350, and to assess their expectations' reflection in the revised version of the standard. We analysed the relationship between stakeholders' beliefs, proposed actions, and the final standard's outcome using PLS-SEM (partial least squares structural equation modelling) in SmartPLS. Furthermore, we examine whether feedback from the accountancy profession, a significant stakeholder group in this process, acts as a moderator of the belief-action relationship. The results confirm that the beliefs significantly influence the proposed actions but have a minimal impact on the outcome, with a small number of suggestions for improvement being integrated. The involvement of the accountancy profession moderated the relationship between beliefs and actions, only for the scope and measurement categories, reflecting their specialised knowledge and technical expertise. Through empirical results, current research contributes to the debate on crypto asset regulation by examining stakeholders' expectations and how their participation has influenced this regulatory field, emphasising the standard-setting body's responsiveness to their input. The study offers theoretical and practical insights into the interplay between stakeholders' beliefs, their proposed actions, and the development of crypto accounting regulation (the outcome).
This study examines whether artificial intelligence (AI) functions as a governance mechanism that reduces the misalignment between firms’ environmental, social, and governance (ESG) disclosures and their underlying ESG performance. Using a sample of Chinese listed firms from 2011 to 2023, we find that AI adoption is associated with lower levels of ESG disclosure-performance misalignment, encompassing both the overstatement and understatement of ESG performance. Cross-sectional analyses further indicate that the association is more pronounced among large firms, non-state-owned enterprises, and firms operating in capital-intensive and technology-driven industries. Additional analysis shows that AI adoption is linked to improved access to green credit, with this relationship mediated by reductions in ESG disclosure-performance misalignment. Overall, this study contributes to the literature on AI and ESG by highlighting AI, particularly learning-based AI, can function as a governance-enabling technology that enhances the alignment between ESG disclosures and actual practices. Our study offers new insights into how different forms of AI (learning-based versus logic-based AI) influence ESG disclosure-performance alignment and capital market responses.
Under China’s policy interventions, the financial shared service center (FSSC), as an innovative application of information technology in corporate accounting systems, has shifted from being primarily market-driven to being jointly promoted by government policies and market forces. Using data from Chinese A-share listed companies from 2003 to 2022 and manually collected FSSC data, this study compares the effectiveness of government promotion versus market forces in driving FSSC implementation. We employ Cox proportional hazards models to investigate the drivers of firms establishing FSSCs. We find that state-owned enterprises (SOEs) establish FSSCs primarily in response to government promotion, whereas non-SOEs are motivated by market competition. Using propensity score matching and multi-period DID models, we find that FSSCs implemented by non-SOEs significantly improve financial reporting quality, whereas those established by SOEs show no measurable impact. These results hold after a series of robustness tests. Furthermore, cross-sectional analysis shows that the effectiveness of FSSCs is particularly pronounced in firms with more subsidiaries, wider geographic dispersion of subsidiaries, stronger subsidiary control, lower earnings management incentives, and adoption of robotic process automation (RPA), with such effects exclusively present in non-SOEs. Our results suggest that market-driven FSSCs generate substantive improvements, while government-promoted implementations fail to yield measurable outcomes. Therefore, when promoting FSSCs, the government needs to pay more attention to firms’ actual conditions and enhance follow-up supervision.
Organizations have been increasingly deploying Robotic Process Automation (RPA) to automate deterministic tasks, especially in accounting and finance. However, prior research highlights a range of RPA-related risks, governance challenges, and a lack of structured approaches to automation. To examine how organizations govern their automation initiatives, we conducted 19 interviews with 24 experienced professionals from 16 large organizations in the United States and Europe. Our study documents RPA governance components, RPA performance measures, the use of unattended and attended bots, and how organizations describe RPA performance in relation to governance. Our findings highlight (1) the important role that the design and implementation of RPA governance play in participants' accounts of successful RPA programs; (2) that performance is portrayed as developing in stages, with improvements in internal process capabilities associated with longer-term organizational performance; (3) most organizations favor the use of unattended bots describing their performance benefits in terms of the availability of suitable processes, cost savings, and stronger control; (4) in contrast, attended automation has been linked to more limited performance benefits that make it less attractive to implement. Our study also provides a structured account of governance practices across the RPA bot lifecycle and of the performance measures used in organizations.
Deep Learning (DL) is a technology with potential to enhance effectiveness and efficiency in audit procedures. However, due to the complexity of DL, it still lacks widespread adoption in the auditing profession. To address this problem, this study follows Design Science Research (DSR) methodology to develop a DL Framework. It integrates the phases of the Cross Industry Standard Process for Data Mining and the phases of risk-based auditing, giving particular attention to guidance for DL. The DL Framework developed here provides detailed application advice on how to augment risk-based auditing with DL. Facing challenges in the realization of our developed DL Framework, we obtained feedback from auditors and computer scientists and identified four key implementation requirements for our DL Framework within organizations. Additionally, we assessed the overall usefulness of our DL Framework based on established DSR evaluation criteria.
Recent advances in artificial intelligence, particularly generative AI (GenAI) and large language models (LLMs), are fundamentally transforming accounting research, creating both opportunities and competitive threats for scholars. This paper proposes a framework that classifies AI-accounting research along two dimensions: research focus (accounting-centric versus AI-centric) and methodological approach (AI-based versus traditional methods). We apply this framework to papers from the IJAIS special issue and recent AI-accounting research published in leading accounting journals to map existing studies and identify research opportunities. Using this same framework, we analyze how accounting researchers can leverage their expertise through strategic positioning and collaboration, revealing where accounting scholars' strengths create the most value. We further examine how GenAI and LLMs transform the research process itself, comparing the capabilities of human researchers and AI agents across the entire research workflow. This analysis reveals that while GenAI democratizes certain research capabilities, it simultaneously intensifies competition by raising expectations for higher-order contributions where human judgment, creativity, and theoretical depth remain valuable. These shifts call for reforming doctoral education to cultivate comparative advantages while building AI fluency.
This paper aims to explore how management accountants craft their roles to meet the demands of IT projects and examines the factors that may hinder this process. It uses the concept of job crafting to operationalize role transition theory and explore micro-level role transitions.Using a multiple case study approach, based on interviews and focus groups involving IT managers and management accountants in 11 organizations, we show that to meet the demands of IT projects, management accountants engage in various forms of job crafting by altering the scope or nature of their tasks, their relationships, and by changing their perception of their work. They also take on additional tasks, create new relationships, and reframe the perception that they have of their job. These job crafting activities take them beyond the roles of bean counter or even business partner. This leads them to act ascontrol architects that engage in promoting collaboration, facilitating knowledge transfer and skill development and creating a decentralized control network.By examining the concrete micro-processes through which management accountants craft their roles, our research enriches the management accounting literature by moving beyond established role dichotomies and introducing the control architect as a novel form of role, while also advancing the role transition literature by showing how transitions unfold at the micro-level through job crafting.
