A categorization of research tasks by perceived necessity and a boilerplate template capturing expected details is suggested, suggesting a categorization of research tasks by perceived necessity and a boilerplate template capturing expected details.
Abstract
As generative AI tools find increasing use in research workflows, ongoing debates on their impact, appropriateness and responsible use have led policymakers to enact policies to disclose AI use at multiple publishing venues. However, are current AI disclosure policies and practices reflective of their purpose? In this work, we first investigate disclosure policies of top computer science venues and find that despite their prevalence, they remain highly under-specified. Secondly, through a survey of computer science researchers (N=$109$), we characterize the necessity of disclosures across different research tasks and levels of human involvement. We learn that researchers find disclosures most necessary for tasks involving research design, and for tasks when the human involvement is low. We also compile expectations that researchers have about the information to be conveyed in AI disclosure statements. Lastly, through an analysis of $13867$ disclosure statements from EMNLP $2025$ and ICLR $2026$, we reveal a large disconnect between these expectations and AI disclosures in practice---a prime example being writing assistance which is deemed less necessary but is frequently disclosed. We conclude with recommendations to align AI disclosure policies and practices with expectations, suggesting a categorization of research tasks by perceived necessity and a boilerplate template capturing expected details.
BACKGROUND
Despite the proliferation of AI disclosure requirements in academic publishing, recent research suggests a persistent gap between policy expectations and research practice. However, little is known about how researchers perceive and navigate these requirements or what limitations they identify in current disclosure practices.
METHOD
This study explored researchers' experiences with AI disclosure through semi-structured interviews with 14 researchers from two interdisciplinary fields, bioinformatics and computational social science. Data were analyzed using reflexive thematic analysis.
RESULTS
Four thematic groupings emerged: fragmented and inconsistently enforced requirements; systemic limitations, including scope ambiguity, research integrity risks, and structural disincentives to honest reporting; researcher perspectives on more effective disclosure practices; and disciplinary variation as a cross-cutting dimension shaping how these issues are experienced across research communities. The findings suggest that the compliance gap reflects an interaction between structural conditions and ethical obligations. This gap is sustained by self-reporting mechanisms that lack verification capacity, a transparency paradox in which honest disclosure can invite professional penalization, and disciplinary norms that resist uniform governance approaches.
CONCLUSIONS
The study provides empirical evidence supporting the development of a structured AI contribution taxonomy as a more principled and practical alternative to existing disclosure practices. More broadly, the findings suggest that effective AI disclosure governance should incorporate field-sensitive adaptation rather than relying on uniform implementation across diverse research communities.
Ayoung Yoon, Siena Oristaglio· Accountability in Research· 0 citations
Generative AI is increasingly used in scholarly research, writing, and publication workflows. Many journal and publisher policies ask authors to disclose relevant AI use, but disclosure alone rarely clarifies how AI-assisted work was performed, what information was entered, how outputs were evaluated, or how human responsibility was maintained. This creates a gap between AI-use disclosure as a publication requirement and AI-use documentation as an open science practice. This article develops a conceptual and practical framework for documenting generative AI use in scholarly workflows. The framework was informed by exploratory, non-systematic source and policy mapping, AI-assisted exploratory evidence mapping, manual review of selected recent literature, and development of accompanying Open Science Framework materials. These steps were used to identify recurring documentation expectations and unresolved policy gaps and to translate them into practical documentation domains, with particular attention to task specificity, proportionality, role-specific documentation, privacy-sensitive transparency, and clinically sensitive contexts. The framework distinguishes disclosure from documentation and proposes documentation fields for minimal and extended AI use. Minimal documentation is intended for low-risk uses such as limited language polishing, whereas extended documentation is recommended when AI supports literature synthesis, coding, analysis, interpretation, manuscript drafting, peer-review-related work, clinical material, or research procedures. The framework is accompanied by reusable OSF materials, including documentation templates, prompt-log structures, declaration examples, checklists, clinical redaction guidance, and source-tracking materials. AI-use disclosure communicates that AI was used; documentation makes the AI-assisted workflow traceable, inspectable, and accountable. A task-specific, proportionate, role-specific, and privacy-sensitive documentation approach can support responsible AI use while protecting confidential, patient-related, peer-review-related, and methodologically sensitive information. The accompanying bilingual materials are openly available on OSF: https://doi.org/10.17605/OSF.IO/A439J.
J. C. Cwik· Research Integrity and Peer...· 0 citations
Generative AI has infiltrated every stage of the research lifecycle: how scholarship is conducted, written, published, and reviewed. Recent policy responses, such as ACM's authorship policy, address an immediate concern about responsible and transparent disclosure of AI use. We argue that a focus on authorship and disclosure, although necessary, risks obscuring and ballooning a set of entrenched problems and strains within publication systems. The central question is not about how papers and other research artifacts should incorporate AI, but how scientific communication itself should evolve when all relevant parties (authors, reviewers, readers) may rely on AI assistance. We draw on our experience within these and other roles to illustrate two contrasting but feasible visions of 2036 with four entwined questions, namely about the purpose of papers as artifacts, reviews, human reviewers, and the incentives that bind all of them. We argue for a shift from policing GenAI and other disruptive technologies to building the infrastructure of provenance, calibration, and accountability that would make trustworthy scholarship the default. We conclude with three grand challenges and invite the community to a broader conversation and research pathways.
The purpose of this study is to examine how the extent, structure and semantic organization of risk disclosures affect the readability of 10-K reports in US financial institutions. The study aims to move beyond traditional keyword-based disclosure measures by introducing artificial intelligence (AI) based linguistic and semantic metrics to better understand how risk transparency interacts with textual clarity.
The study combines traditional and AI-driven textual analysis methods to measure both risk disclosure and readability. Risk disclosure is captured using dictionary-based keyword measures as well as AI-based topic modeling and semantic embedding techniques that quantify thematic dispersion, concentration, dominance and novelty. Readability is measured using the Bog Index together with AI-based linguistic indicators of syntactic complexity and semantic atypicality. The empirical analysis is conducted on 4,649 firm-year observations of US banks and insurance companies over the period 2005–2023.
The results show that greater risk disclosure is consistently associated with lower readability across both traditional and AI-based measures. Higher risk-term intensity, broader thematic dispersion and greater semantic novelty increase linguistic and structural complexity, making reports harder to read. In contrast, concentrated and semantically cohesive risk narratives improve readability. AI-based topic and embedding measures produce stronger and more consistent effects than keyword counts alone. Additional analyses indicate that positive disclosure tone, stronger board effectiveness and higher auditor effectiveness mitigate the negative readability effects of extensive risk reporting. Crisis conditions, such as the COVID-19 period, amplify disclosure dispersion and further reduce readability.
To the best of the authors’ knowledge, this study is among the first to jointly integrate dictionary-based, topic-model, syntactic and semantic embedding approaches to analyze the relationship between risk disclosure and report readability. It demonstrates that AI-based linguistic measures add substantial explanatory power beyond traditional readability and keyword metrics and offers a comprehensive framework for evaluating disclosure clarity in financial reporting.
Zabihollah Rezaee, Javad Rajabalizadeh· Journal of Financial Reporti...· 0 citations
The use of generative artificial intelligence (AI) in scholarly publishing is expanding rapidly, yet clear standards for its appropriate use and disclosure remain lacking. Surveys indicate that many researchers already use AI tools for manuscript preparation, particularly for writing assistance and error detection. However, attitudes toward acceptable AI use and disclosure requirements remain inconsistent, especially regarding the use of AI in drafting manuscripts and in the peer-review process. Analyses of manuscript submissions to major journal groups suggest that the proportion of authors disclosing AI use is substantially lower than estimates from researcher surveys, indicating possible underreporting or uncertainty about reporting requirements. It is therefore essential to implement standardized frameworks that require explicit disclosure of AI use to ensure transparency and uphold trust in scholarly communication.
D. Suh, D. Han· Journal of Korean medical sc...· 2 citations
Generative AI is rapidly reshaping Open Source Software (OSS) software development,prompting projects to introduce policies governing AI-assisted contributions. However, little is known about how these policies differ or whether they influence developer experience. We present the first large-scale empirical study of AI governance policies in OSS. Analyzing 29,624 GitHub repositories, we identify 385 projects that adopted AI policies and derive TRACE, a framework capturing five governance dimensions: Transparency, Responsibility, Attribution, Constraints, and Enforcement. We further classify policies into five governance families and estimate their effects using propensity-score matching and longitudinal difference-in-differences analysis. Our results show that AI governance primarily regulates rather than prohibits AI-assisted development. Policy adoption brings maintainer engagement, increased AI disclosure, richer review interactions, and improved code quality while AI-assisted contributions continue to grow. Governance design matters: policies emphasizing transparency and responsibility produced stronger community and quality outcomes than restrictive approaches alone. Our findings show how different AI governance strategies shape developer experience and provide evidence to help OSS communities design effective AI policies.
Yunqi Chen, Thomas Zimmermann, Bianca Trinkenreich· 0 citations
A new method, called CW-Net, translates the reasoning process of an autonomous vehicle’s AI system into understandable concepts that explain its behavior.