Category A — Original papers
UAI-style formatting, up to 4 pages (no limit on supplementary material).
Machine Learning/CORE Unranked
SafeAI 2026
Second Workshop on Safe AI
SafeAI@UAI 2026 is the second workshop focused on safety challenges in agentic AI — systems that autonomously act, adapt, and interact over extended horizons. Held in Amsterdam as part of UAI 2026, the workshop features invited talks, a panel discussion, and contributed papers on foundations, robustness, interpretability, and deployment of safe AI systems, with no formal proceedings to allow dual submission.
Paper fit
Contribution paths
A strong submission should clearly identify its contribution and evaluate it appropriately.
Category B — Recently accepted and under-review papers
Submitted in the original format and length from top ML/AI conferences where they were accepted or are under review.
Research areas in scope
01
Topics of interest
Foundations for safe and agentic AIFoundations of Safe AI across learning paradigms, including reinforcement and continual learningSequential decision-making under uncertainty, including Bayesian decision theory and POMDPsObjective specification, reward design, and alignment for agentic systemsTest-time scaling, adaptive inference, and safety implications of increasingly capable agentsUncertainty, robustness, and controlUncertainty quantification, calibration, and propagation in decision-makingRisk-sensitive planning and long-horizon safety under uncertaintyRobustness to distribution shift, adversarial conditions, and non-stationary environmentsSafety and control in RL-based agents, including safe exploration and learning-based controlInterpretability, auditing, and evaluationInterpretability and explainability of agent policies, representations, and behaviorsMethods to audit, measure, monitor, and evaluate agentic AI systemsBenchmarks, metrics, and case studies for safe and aligned agentsAdversarial evaluation and stress testing of agentic systems, including red teamingEngineering, deployment, and applicationsAgent failure modes: reward hacking, goal misgeneralization, agent hijacking, and emergent behaviorsHuman-in-the-loop and oversight mechanisms for agentic systemsDeployment-time safety of AI in safety-critical and autonomous systems
Policies worth checking twice
- Single-blind peer review by the program committee
- At least one author must attend in person
- No proceedings will be published, allowing authors to submit elsewhere
- Submissions under review at other venues are allowed, provided they do not breach dual-submission or anonymity policies of those venues
- Accepted submissions will be made publicly visible on the workshop site and OpenReview
- Supplementary material may be included in the same PDF after references, but reviewers may choose whether to consult it
- Papers must comply with UAI style requirements using the adjusted SafeAI template for original submissions
Official sources
Compiled from the official call for papers. The organizers’ pages remain authoritative.
Last verified September 9, 2026