Full papers
Up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option).
Machine Learning/CORE Unranked
Child Safety in AI Workshop 2026
NeurIPS 2026 Workshop on Child Safety in AI
The NeurIPS 2026 Workshop on Child Safety in AI brings together researchers, practitioners, and policymakers to address technical and sociotechnical challenges in protecting children within AI ecosystems. It focuses on mitigating risks such as developmental harm, harmful content generation, and exploitation, while navigating ethical constraints like restricted data access and the need for cross-sector collaboration.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Works-in-progress
Up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option).
Position papers
Up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option).
Open-problem submissions
Up to four pages, excluding references, using the NeurIPS 2026 template (with the 'dblblindworkshop' option).
Research areas in scope
01
Safe Data, Evaluation, and Benchmarking
Safe data curation and governance for child-related dataEvaluation methodologies under restricted access settings (e.g., proxy tasks, synthetic benchmarks)Measurement and benchmarking of child safety risks (e.g., datasets, metrics, eval protocols, long-term evals)Auditing and interpretability methods for detecting unsafe capabilitiesData-free or privacy-preserving auditing techniques
02
Robust and Safe Model Design
Preventing the emergence of harmful capabilities (e.g., defenses against jailbreaking, resilience to harmful fine-tuning, preventing concept fusion, data cleaning)Evaluating capability degradation when implementing safety solutionsAdversarial robustness and red teaming for child safetyMachine unlearning and concept erasure with strong guaranteesChild safety in multimodal and agentic AI systems
03
Deployment, Monitoring, and Ecosystem Safeguards
Robust safeguards for model deployment (e.g., input/output filtering, monitoring, open-weight model defenses)Content provenance, watermarking, and traceabilitySafety–privacy trade-offs in AI systemsProtection of user-generated content from manipulationCross-platform and ecosystem-level safety coordinationHuman factors and moderator well-being
04
Human-Centered Design, Policy, and Societal Implications
Child-centered safety design and age-appropriate interfacesPolicy, governance, and regulatory frameworksCollaboration with external stakeholders (NGOs, law enforcement, hotlines)Ethical, legal, and societal implications, including global perspectives
Policies worth checking twice
- Submissions must be anonymous (double-blind review)
- Submissions are limited to four pages, excluding references
- An optional unlimited-length appendix is permitted
- Accepted papers are non-archival and may be submitted to other venues
- Evaluation prioritizes potential to stimulate productive workshop discussion alongside technical soundness
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026