Short paper
Max 4 pages, excluding references; can include unlimited appendix within the same PDF as long as main text stays within limit.
Machine Learning/CORE Unranked
ICLR 2026
ICLR 2026 Workshop on Principled Design for Trustworthy AI - Interpretability, Robustness, and Safety across Modalities
The ICLR 2026 Workshop on Principled Design for Trustworthy AI focuses on developing interpretable, robust, and safe AI systems across modalities such as language, vision, audio, and time series. It addresses critical challenges in high-stakes applications like healthcare and autonomous driving by unifying research on interpretability, robustness, uncertainty, and safety throughout the AI lifecycle.
Key deadlines
Verified September 9, 2026Abstract registration
February 3, 2026
AoE
Full paper
February 3, 2026
AoE
Workshop timeline
Submission and decisions
Abstract registrationKey deadline
February 3, 2026 · AoE
Full paperKey deadline
February 3, 2026 · AoE
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Long paper
Max 9 pages, excluding references; can include unlimited appendix within the same PDF as long as main text stays within limit.
Research areas in scope
01
Interpretable and Intervenable Models
concept bottlenecks and modular architecturesmechanistic interpretability and concept-based reasoninginterpretability for control and real-time intervention
02
Inference-Time Safety and Monitoring
reasoning trace auditing in LLMs and VLMsinference-time safeguards and safety mechanismschain-of-thought consistency and hallucination detectionreal-time monitoring and failure intervention mechanisms
03
Multimodal Trust Challenges
grounding failures and cross-modal misalignmentsafety in vision-language and deep vision systemscross-modal alignment and robust multimodal reasoningtrust and uncertainty in video, audio, and time-series models
04
Robustness and Threat Models
adversarial attacks and defensesrobustness to distributional, conceptual, and cascading shiftsformal verification methods and safety guaranteesrobustness under streaming, online, or low-resource conditions
05
Trust Evaluation and Responsible Deployment
human-AI trust calibrationconfidence estimationuncertainty quantificationmetrics for interpretability/alignment/robustnesstransparent and accountable deployment pipelinessafety alignment
06
Safety and Trustworthiness in LLM Agents
safety and failures in planning and action executionemergent behaviors in multi-agent interactionsintervention and control in agent loopsalignment of long-horizon goals with user intentauditing and debugging LLM agents in real-world deployment
Policies worth checking twice
- Reviews are double-blind.
- Accepted papers are non-archival.
- Submissions must be original and not accepted in other archival venues by the submission deadline; violation results in desk rejection.
- New OpenReview profiles without institutional email require up to two weeks for moderation.
- New OpenReview profiles with institutional email are activated automatically.
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026