Expectations and Practices around AI Disclosure in CS Research

Arati Mohapatra; Danish Pruthi

cs.CYAug 24, 2026

Expectations and Practices around AI Disclosure in CS Research

Authors: Arati Mohapatra, Danish Pruthi

Organizations: Indian Institute of Science Bengaluru, KA, India

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.

Figures & tables

Figure 1: ACM SIGCOMM’s AI Disclosure policy.

Level	Description
Assumed	No explicit mention of human involvement. Participants assume a certain baseline level of human involvement.
High	When the author provides most of the contribution and initiative, while AI outputs are closely supervised and verified.
Low	When the AI assistant provides most of the contribution and initiative, and AI outputs are loosely supervised and verified by the author.

Table 1: Descriptions of different levels of human involvement considered in the survey (See § 5.1 ).

Dimension	Variants
Detail	Task
	Model name
	Reason for AI use
	Non-use of AI
	Human oversight
	Responsibility declaration

Table 2: Dimensions along which survey participants indicate their preferences for framing AI disclosures.

Figure 2: AI disclosure necessity ratings across research phases and human involvement (HI) conditions.

Panel of Table 3 — Table 3: Research tasks ordered by mean disclosure-necessity rating. The third column reports the necessity of disclosure, by identifying the nearest option in the Likert scale, and the Mean Rating column visualizes each task’s mean and 95% confidence interval. Significance markers ( ${}^{*}p<.05$ ) indicate differences from the median-rated task ( Suggest experimental parameters ) in a linear mixed-effects model.

Figure 3: Expected vs. observed coverage of disclosure details, comparing survey expectations against ICLR and EMNLP practice.

Panel of Figure 4 — Figure 4: An identical AI disclosure statement appearing in 95 ICLR 2026 submissions.

Panel of Figure 5 — Figure 5: A structured AI disclosure template, designed to capture details that survey respondents prioritize.

Research Tasks by Disclosure Level
Mandatory (rating $\geq 3.5$ )
Generate synthetic data sets
Develop theoretical models or frameworks
Propose new hypotheses
Recommended ( $2.5\leq$ rating $<3.5$ )
Design research methodology or experiments

Table 4: Suggested disclosure levels for research tasks, based on ratings from survey participants.

Appendix figures & tables19 assets

Supplementary material from the paper’s appendix.

Appendix

Area	Conference	n	Year	Disclosure
AI	NeurIPS, ACL, EMNLP, SIGIR, WWW	5	2026	Yes
	ICLR, NAACL	2	2025	Yes
	AAAI, IJCAI, CVPR, ECCV, ICML	5	2026	No
	ICCV	1	2025	No
Systems	ISCA, SIGCOMM, NSDI, CCS, IEEE S&P, SIGMOD, DAC, RTAS, HPDC, ICS, MobiSys, IMC, SIGMETRICS, FSE, ICSE	15	2026	Yes
	RTSS, SC	2	2025	Yes

Appendix

Table 5: AI disclosure policy presence across major CS conferences.

Society	Conferences	n	#Changes
AAAI	–	0	1
ACL	ACL, EMNLP, NAACL	3	2
ACM	SIGIR, WWW, SIGCOMM, CCS, SIGMOD, DAC, HPDC, ICS, MobiSys, IMC, SIGMETRICS, FSE, ICSE, STOC, CRYPTO, EuroCrypt, SIGGRAPH, SIGGRAPH Asia, SIGCSE, CHI, ISCA	21	1
IEEE	ISCA, IEEE S&P, ICSE, ICRA, IROS, VIS, VR	7	4
None	ICLR, NeurIPS, NSDI, RTAS, RTSS, SC	6	–

Appendix

Table 6: Conferences inspired by each society’s disclosure policy, and the number of revisions made.

Phase	Task
Idea Generation	Discover research topics or identify gaps in current research
	Identify relevant literature
	Summarize or analyse existing literature
	Propose new hypotheses
Research Design	Help design research methodology or experiments
	Help develop theoretical models or conceptual frameworks

Appendix

Table 7: Taxonomy of research phases and associated tasks.

Detail	Description	Example
Task	The specific task AI was used for.	“In writing this work, LLMs have been used in helping find relevant works, format LaTeX, and implement code, and proofread. ”
Model name	The AI model or system that was used.	“Specifically, we employed Claude Sonnet 4 (Anthropic) and GPT-5 (OpenAI) for language polishing and refinement purposes.”
Reason	The reason AI was used instead of a non-AI method.	“We use large language models (LLMs) to support labor-intensive and mistake-prone work. ”
Non-use	Tasks for which AI was not used	“ LLMs were not used for generating novel scientific ideas, experiments, or analyses. ”
Human oversight	The extent of human oversight or contributions during or after AI use.	“All outputs from LLMs were carefully reviewed, verified, and edited by the authors to ensure correctness and originality.”
Responsibility	The authors’ declaration of responsibility for outcomes of AI use, such as potential errors or biases.	“The authors retain full responsibility for the entire content, including any errors or inaccuracies. ”

Appendix

Table 8: Disclosure details, with the relevant portion of each example highlighted.

Panel of Table 9 — Table 9: Welcome screen of the survey instrument.

Panel of Table 10 — Table 10: Introduction and Consent section of the survey instrument.

Panel of Table 11 — Table 11: Presence and Framing of AI Disclosure Statements section of the survey instrument.

Panel of Table 12 — Table 12: Importance of AI Disclosure in Idea Generation section of the survey instrument. All subsequent phases follow the same template, and are thus excluded for brevity.

Panel of Table 13 — Table 13: Additional Views on AI Disclosure Necessity section of the survey instrument.

Panel of Table 14 — Table 14: Demographics section of the survey instrument.

Demographic	Level	$n$
Gender	Female	$36$
	Male	$67$
	Non-binary	$1$
	Prefer not to disclose	$5$
Country of work / residence	India	$46$
	United States	$40$

Appendix

Table 15: Demographics of survey respondents (

N=109

). Mean age:

26.64

years (SD =

6.28

). Counts in the

n

column are unweighted respondent totals; categories within each demographic are mutually exclusive.

Panel of Table 16 — Table 16: Prompt used for extracting ICLR disclosures.

Panel of Table 17 — Table 17: First-stage prompt for the two-stage EMNLP extraction pipeline. Reads the Responsible NLP Checklist (Section E and E1) and returns AI-use status, disclosure type (inline vs. paper-section pointer), and the corresponding content or section reference.

Panel of Table 18 — Table 18: Second-stage prompt for the two-stage EMNLP extraction pipeline. Reads the referenced section of the paper PDF and returns whether a genuine AI usage disclosure is present, the section heading, and the disclosure text itself.

Annotator	Details			Tasks			Length Acc.
Annotator	P	R	F1	P	R	F1	Length Acc.
Annotator $1$	$98.2$	$97.4$	$97.8$	$91.8$	$93.7$	$92.7$	$90.0$
Annotator $2$	$98.2$	$95.7$	$96.9$	$88.4$	$90.2$	$89.3$	$86.0$
Annotator $3$	$91.1$	$98.6$	$94.7$	$91.1$	$88.7$	$89.9$	$87.0$
Average	95.8	97.2	96.5	90.4	90.9	90.6	87.7

Appendix

Table 19: Validation of LLM-based annotation against

3

human annotators on

N=100

disclosures. We report micro precision (P), recall (R), and F1 for the multi-label Details and Tasks schemas, and classification accuracy for the single-label Length schema. All values are percentages; numbers compare each human annotator’s labels to Gemini’s annotations.

Annotator A	Annotator B	Details			Tasks			Length Acc.
Annotator A	Annotator B	P	R	F1	P	R	F1	Length Acc.
Annotator $1$	Annotator $2$	$94.8$	$96.5$	$95.6$	$90.2$	$90.2$	$90.2$	$81.0$
Annotator $1$	Annotator $3$	$98.1$	$89.9$	$93.8$	$87.3$	$91.6$	$89.4$	$90.0$
Annotator $2$	Annotator $3$	$98.6$	$88.7$	$93.4$	$84.7$	$88.8$	$86.7$	$80.0$
Average		97.2	91.7	94.3	87.4	90.2	88.8	83.7

Appendix

Table 20: Pairwise inter-annotator agreement on

N=100

Panel of Table 21 — Table 21: Annotation prompt ( $1$ / $3$ ): defines the seven detail labels ( task , model , reason , non_use , human_oversight , responsibility , purpose ) and the disambiguation rules.

Panel of Table 22 — Table 22: Annotation prompt ( $2$ / $3$ ): defines the task labels (e.g., edit_clarity , draft_paper , software_code , synthetic_data ) and disambiguation rules for borderline cases.

Panel of Table 23 — Table 23: Annotation prompt ( $3$ / $3$ ): defines the length categories, general output rules, and worked examples for the model to follow.

AI Use Disclosure

See all 10 papers

Jun 9, 2026

Designed by Journalists, but Is It for Readers? Rethinking AI Disclosures and Transparency in News

As newsrooms integrate generative AI, journalists face a disclosure challenge: how to communicate AI involvement in ways that maintain reader trust. Current practice offers two approaches: brief one-line labels or detailed disclosures specifying human oversight, editorial accountability, and error reporting mechanisms. Neither achieves journalists' goal of building trust through transparency. An existing controlled experiment with 34 news readers show that detailed disclosures trigger a transparency dilemma, reducing trust rather than increasing it, and risk introducing dark patterns that readers scroll past with the illusion of transparency. One-line disclosures avoid this effect but can create an information gap, prompting readers to expend cognitive effort searching for signs of AI involvement that the disclosure indicates but does not explain. Yet readers are not rejecting transparency, they proposed disclosure designs centered on user agency: detail-on-demand interactions, proportional AI-ratio visualizations, outlet-level signals, and explicit "no AI" labels. I argue that this disconnect between what practitioners believe is responsible disclosure and what users actually need is a design problem for the HCI community.

Pooja Prajod

Centrum Wiskunde & Informatica Amsterdam, the Netherlands

Human-AI Interaction Trust in AI AI Use Disclosure

Ask PDF

Apr 13, 2026

Inspectable AI for Science: A Research Object Approach to Generative AI Governance

This paper introduces AI as a Research Object (AI-RO), a paradigm for governing the use of generative AI in scientific research. Instead of debating whether AI is an author or merely a tool, we propose treating AI interactions as structured, inspectable components of the research process. Under this view, the legitimacy of an AI-assisted scientific paper depends on how model use is integrated into the workflow, documented, and made accountable. Drawing on Research Object theory and FAIR principles, we propose a framework for recording model configuration, prompts, and outputs through interaction logs and metadata packaging. These properties are particularly consequential in security and privacy (S&P) research, where provenance artifacts must satisfy confidentiality constraints, integrity guarantees, and auditability requirements that generic disclosure practices do not address. We implement a lightweight writing pipeline in which a language model synthesizes human-authored structured literature review notes under explicit constraints and produces a verifiable provenance record. We present this work as a position supported by an initial demonstrative workflow, arguing that governance of generative AI in science can be implemented as structured documentation, controlled disclosure, and integrity-preserving provenance capture. Based on this example, we outline and motivate a set of necessary future developments required to make such practices practical and widely adoptable.

Ruta Binkyte, Sharif Abuaddba, Chamikara Mahawaga +3

CISPA, Helmholtz Center for Information Security · DATA61,CSIRO · Data61, CSIRO +1

AI Governance AI-Assisted Scientific Research AI Accountability Data Provenance

Ask PDF

Jun 3, 2025

Position: EU AI Act's Research Exemptions Can Break the Publication Norms of Major AI Conferences

The EU has become one of the vanguards in regulating the digital age. A particularly important regulation in the Artificial Intelligence (AI) domain is the 2024 enacted EU AI Act. The AI Act specifies -- due to a risk-based approach -- various obligations for providers of AI systems. These obligations, for example, include a cascade of documentation and compliance measures, which represent a potential obstacle to science. But do these obligations also apply to AI researchers? This position paper argues that, indeed, the AI Act's obligations could apply in many more cases than the AI community is aware of. Moreover, we argue that the AI Act is drafted in a manner that may unwillingly disrupt the scientific publication practices of the AI research community, with a focus on model and system release. We contribute the following: 1. We offer a high-level roadmap for AI researchers to evaluate whether they need to comply with the AI Act 2. We explain with everyday research examples why the AI Act applies to AI research. 3. We analyse the exceptions of the AI Act's applicability AI research and offer visual tool for researchers to navigate the AI Act's complex system or research exceptions 4. We establish a position the AI Act's research exceptions fail to account for current AI research conventions, as publishing AI research may void the research exceptions of the Act. 5. We propose changes to the AI Act to provide more legal certainty for AI researchers and give two recommendations for AI researchers to reduce the risk of not complying with the AI Act. We see our paper as a starting point for a discussion between policymakers, legal scholars, and AI researchers to avoid unintended side effects of the AI Act.

Alina Wernick, Kristof Meding

CZS Institute for AI and Law, University Tübingen, Germany.

AI Governance

Ask PDF

Abstract

AI Use Disclosure

Explore similar work

Designed by Journalists, but Is It for Readers? Rethinking AI Disclosures and Transparency in News

Inspectable AI for Science: A Research Object Approach to Generative AI Governance

Position: EU AI Act's Research Exemptions Can Break the Publication Norms of Major AI Conferences