Policy on the Use of Generative AI for Research and Writing Papers for Authors

Generative AI tools, including large language models, are increasingly used throughout the research process. ICML permits the use of these tools in preparing submissions, provided that their use is appropriately disclosed and that the authors remain fully responsible for all aspects of the submitted work.

Required AI Disclosure Statement

All submissions must include an AI Disclosure Statement at the end of the paper, before or after the references.

The AI Disclosure Statement does not count toward the paper page limit.

Authors must disclose any substantive use of generative AI tools in conducting the research or preparing the submission. The disclosure should describe in detail

1. how generative AI tools (e.g., LLMs) were used, including the research or writing tasks for which they were used
2. how AI assisted outputs were evaluated or verified, where applicable
3. any use of generative AI that is material to understanding, reproducing, or evaluating the work

Authors do not need to provide complete prompts or transcripts unless these are necessary for reproducibility or form part of the research methodology.

Uses Requiring Disclosure

Authors must disclose the use of generative AI tools for substantive research tasks. These include, but are not limited to

1. generating or substantially developing research ideas, hypotheses, or research questions
2. developing theoretical models or conceptual frameworks
3. formulating mathematical claims, deriving results, or assisting with proofs
4. designing or substantially modifying research methodology or experiments
5. implementing methods or generating substantial portions of experimental code
6. generating, cleaning, transforming, annotating, or analyzing data
7. generating synthetic data used in the research
8. interpreting experimental or theoretical results
9. conducting or synthesizing literature reviews when the output materially informs the paper
10. generating or substantially modifying scientific figures, tables, or other research artifacts
11. drafting substantial portions of the manuscript or translating substantive scientific content

Authors should err on the side of disclosure when it is unclear whether a particular use is substantive.

Generative AI as Part of the Research Methodology

When generative AI is used as part of the research methodology, for example to generate or annotate data, evaluate model outputs, produce synthetic data, act as a component of an experimental pipeline, or otherwise contribute directly to the reported results, its use must be described in the main paper with sufficient detail to allow readers to understand and, where appropriate, reproduce the methodology.

The AI Disclosure Statement is intended to describe the authors’ use of generative AI in conducting or preparing the research. It does not replace methodological reporting required to understand or reproduce the scientific work.

Uses That Do Not Require Disclosure

Routine editorial assistance does not need to be disclosed. This includes uses limited to

1. correcting spelling or grammar
2. improving sentence level clarity or readability without changing scientific content
3. formatting text, equations, tables, references, or LaTeX
4. providing autocomplete or other routine editing assistance

If such tools materially alter the intellectual or scientific content of the paper, their use should instead be disclosed.

Author Responsibility

The authors are fully responsible for all content in their submission, regardless of whether generative AI tools were used in its preparation.

This includes responsibility for the correctness and originality of scientific claims and conclusions, mathematical arguments and proofs, experimental results, code and data, citations and references, text, figures, and other research artifacts.

Generative AI systems may not be listed as authors.

Authors are responsible for verifying AI generated or AI assisted material and for ensuring that the submission complies with ICML policies on research integrity, plagiarism, attribution, confidentiality, and intellectual property.

A substantial falsehood, fabricated citation, plagiarism, misrepresentation, or other research integrity violation remains the responsibility of the authors even when it originated from a generative AI system.

Disclosure and Enforcement

Failure to disclose substantive generative AI use, or materially misrepresenting how such tools were used, may constitute a violation of ICML research integrity policies and may result in action by the Program Chairs, including desk rejection in serious cases.

The purpose of the disclosure is transparency, not to penalize responsible use of generative AI.

The fact that generative AI tools were used should not, by itself, be treated positively or negatively in the evaluation of a submission. Reviewers should evaluate the scientific contribution of the work according to the usual ICML criteria.

Example AI Disclosure Statements

The following examples are illustrative. Authors should adapt the statement so that it accurately reflects how generative AI tools were used in their work.

General example

AI Disclosure Statement. We used generative AI tools to assist with [describe tasks, for example debugging experimental code, improving the presentation of several proofs, or editing the manuscript for clarity]. All AI assisted outputs were reviewed and verified by the authors. Where appropriate, the authors independently checked the relevant mathematical claims, code, experimental results, and citations. The authors take full responsibility for the content of the paper, including all claims, results, code, and references.

Example 1. No substantive AI use

AI Disclosure Statement. Generative AI tools were not used for substantive research tasks or for generating scientific content in this work. The authors may have used routine spelling, grammar, or formatting assistance that falls within the editorial use exemption described above.

Example 2. Writing and presentation assistance

AI Disclosure Statement. The authors used generative AI tools to improve the clarity, grammar, and presentation of portions of the manuscript. The tools were not used to generate research ideas, scientific claims, mathematical arguments, experimental results, or citations. All suggested edits were reviewed and approved by the authors, who take full responsibility for the final text.

Example 3. Literature search and related work

AI Disclosure Statement. Generative AI tools were used to assist with identifying potentially relevant literature and organizing themes for the related work discussion. The authors independently located and read the cited sources and verified all bibliographic information and claims attributed to prior work. AI generated summaries were not treated as authoritative sources.

Example 4. Code generation and debugging

AI Disclosure Statement. The authors used generative AI tools to assist with writing and debugging portions of the experimental code, including implementation support, data processing utilities, and debugging suggestions. All AI generated code was reviewed, tested, and modified by the authors as necessary. The authors verified the correctness of the final implementation and take full responsibility for the reported experimental results.

Example 5. Mathematical derivations or proofs

AI Disclosure Statement. Generative AI tools were used during the development of several mathematical derivations and to suggest possible intermediate proof steps. The final statements, derivations, and proofs were independently checked and written by the authors. The authors verified all mathematical arguments and take full responsibility for their correctness.

Example 6. Research ideation and experimental design

AI Disclosure Statement. The authors used generative AI tools during early stage brainstorming to explore possible research questions, alternative formulations of the problem, and potential experimental controls. The final research questions, methodology, theoretical claims, and experimental design were developed and validated by the authors. AI generated suggestions were treated as exploratory input rather than as scientific evidence.

Example 7. Multiple substantive uses

AI Disclosure Statement. The authors used generative AI tools for several tasks during this project, including literature exploration, debugging experimental code, suggesting alternative presentations of mathematical arguments, and improving the clarity of portions of the manuscript. All generated outputs were independently reviewed and verified by the authors, including all code, mathematical claims, scientific statements, and citations. The authors take full responsibility for the content of the paper.

Example 8. Generative AI as part of the research methodology

AI Disclosure Statement. A generative AI system was used as part of the research methodology reported in this work, including for data annotation, synthetic data generation, automated evaluation, or another component of the experimental pipeline. The model, prompting procedure, inference settings, validation procedures, and role of the generated outputs are described in the main paper or supplementary material as appropriate. The authors reviewed and validated the resulting outputs and take full responsibility for the conclusions drawn from them. The disclosure statement is intended to provide transparency and does not replace the methodological description required to understand and reproduce the work.