LLM / AI-Assistance Disclosure Policies (2026-06)

Factual reference for where and whether AI assistance must be disclosed at major venues. This file describes disclosure obligations only. It does not provide any advice on evading AI-detection — the deai module is a readability/clarity aid, not a detector-evasion tool. When a paper used an LLM in any non-trivial way, disclose it per the target venue's policy below.

All listed venues permit language polishing. They differ on what must be disclosed and where. "Polishing" = grammar/spelling/wording; "generative use" = drafting text, generating code, or producing analysis.

Venue / publisher	Polishing	Generative use disclosure	Where	Desk-reject risk
NeurIPS 2025/2026	exempt	declare in the paper checklist when an LLM is a core method component	checklist	checklist required; LLMs cannot be authors
ICML 2026	exempt	author fully responsible; prompt injection forbidden	n/a	prompt injection = desk reject
ICLR 2026	exempt	significant ideation/writing role must be described	dedicated LLM-usage section (may be appendix)	undisclosed significant use → desk reject
COLM 2026	exempt (light aid)	significant role must be disclosed	dedicated section (ICLR-style)	undisclosed significant use → desk reject
ACL / ARR	exempt	generative writing/code must be declared	Responsible NLP Checklist + Acknowledgements	wrong/misleading checklist → desk reject
CVPR	n/a (no mechanism)	author fully responsible	n/a	fabricated citations / factual errors → reject without review
IEEE	"suggested, not required"	AI-generated content (text/figures/code) must be disclosed	Acknowledgments section (name system + affected sections)	—
ACM	exempt after revision	generated content disclosed prominently	within the Work (e.g. Acknowledgements)	LLMs cannot be authors
Springer Nature	"AI-assisted copy editing" exempt	generative use recorded	Methods	AI-generated images forbidden; LLM not an author
Elsevier	grammar/spelling exempt	all other generative use declared	a "Declaration of Generative AI…" statement before the references	published with the paper
Science	—	AI use disclosed	cover letter and in the paper	undisclosed use can be misconduct
ICMJE journals (medical)	—	AI use disclosed	cover letter and manuscript (Section V, 2026-01)	undisclosed use can constitute misconduct

Practical guidance

• arXiv: generative AI cannot be listed as an author; significant use should be disclosed per field convention. Since 2025-10-31, arXiv CS review/survey/position articles need proof of peer review at a journal/conference (workshops do not count).
• A cover letter is the right place for disclosure at medical/ICMJE journals, Science, and many Elsevier titles; ML conferences use checklists or a dedicated section instead. See cover-letter skill for the letter-side declaration.
• When in doubt, disclose. Disclosure is never penalized; undisclosed significant use is, at several venues, grounds for desk rejection or a misconduct finding.

Sources: per-venue 2026 author guides and CFPs; ICMJE 2026-01 Recommendations; IEEE / ACM / Springer Nature / Elsevier AI policy pages; arXiv help/blog. See the parent audit's latex-paper-en-venue-factcheck.md for the full citation list.