Main track papers
3-6 page submissions on attributing model behaviors, including experiments and technical contributions.
Machine Learning/CORE Unranked
ATTRIB 2026
Third NeurIPS Workshop on Attributing Model Behavior at Scale: Data Attribution and Provenance
ATTRIB 2026 is a workshop co-located with NeurIPS 2026 in Sydney, Australia, focused on advancing the field of attributing model behaviors to training data. It brings together researchers and practitioners to address technical challenges in contributive and corroborative attribution methods and their real-world applications in law, journalism, AI safety, and beyond, with an emphasis on making attribution interpretable, auditable, and practically useful.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Idea track papers
2-4 page submissions including vision papers, unifying ideas, or documentation of failed experiments; may be opinionated without experiments, provided opinions are clearly demarcated.
Research areas in scope
01
Attribution Methods
Contributive attribution: reliability, robustness, and verification of causal effects of training dataCorroborative attribution: evaluation, verification, and meaningfulness for downstream usesAttribution of synthetic data: implications for copyright, provenance, and liability
02
Downstream Applications
Data attribution for accountability and trustworthy AIApplications in law, journalism, music, and regulatory auditsAttribution in the context of model-generated data and blurred data boundaries
Policies worth checking twice
- Submissions must be double-blind (use 'dblblindworkshop' template)
- References and appendices are not included in page limits
- Appendix is included in the same PDF as the main body with no page limit
- Reciprocal reviewing model: at least one author must serve as a reviewer for another submission
- Submissions without a designated reciprocal reviewer may be desk rejected
- LLM usage policy follows NeurIPS 2026 guidelines
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026