Full papers
Submissions may use up to eight pages, excluding references and appendices, and must include: problem specification, proposed approach, observed outcome, and reason for failure, with quantitative evidence and diagnostics.
Machine Learning/CORE Unranked
ICBINB-BIO 2026
I Can’t Believe It’s Not Better (ICBINB): Failure Modes of AI in Biology
ICBINB-BIO 2026 is a NeurIPS workshop focused on identifying and analyzing failure modes of AI systems in biological applications. It brings together machine learning and life science researchers to study why AI models break under real-world conditions such as distribution shift, weak supervision, and deployment constraints, with the goal of building more reliable and trustworthy biological AI systems.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Tiny papers
Submissions may use up to four pages of main text; must state the problem and provide evidence of at least one negative or unexpected outcome, without requiring a full causal analysis.
Research areas in scope
01
Topics of Interest
Out-of-distribution generalization and domain shiftFailure under weak or confounded supervisionCausal mechanisms versus spurious correlationUncertainty, calibration, and decision-aware reliabilityInterpretability and trustworthy biological inferenceLearning with limited data and distribution shiftDeployment-relevant evaluation beyond benchmarksLimits of foundation, multimodal, and agentic models for biologyCausal intervention and experimental design
Policies worth checking twice
- Submissions are double-blind; linked material must preserve anonymity.
- Authors may include unlimited appendices, but reviewers are not required to read them.
- Disclose any use of large language models (LLMs) with a short paragraph describing their role.
- Ethics and reproducibility statements are recommended but do not count toward the page limit.
- Concurrent submissions are welcome if they comply with the target venue’s policy.
- Submissions must not have been accepted for publication in previous conference proceedings.
- Work concurrently under review for NeurIPS 2026 main track is welcome.
- Accepted papers will appear on OpenReview and the workshop remains non-archival.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026