Statistical Frameworks for Uncertainty in Agentic Systems

Topics

We invite submissions on the foundations of uncertainty in agentic systems. We welcome theoretical, methodological, and empirical contributions, as well as benchmarks and application-driven case studies.

Topics of interest include, but are not limited to:

• Conformal prediction and distribution-free risk control for LLM or agent outputs
• Calibration and uncertainty reporting for generative models
• Selective prediction, abstention, and hallucination mitigation
• Sequential monitoring, evidence aggregation, and stopping rules
• Prediction-powered inference with noisy or machine-generated labels
• Uncertainty under prompt variation, adaptivity, and distribution shift
• Auditing and verification for tool use and external feedback loops
• Compositional guarantees across multi-stage or multi-agent workflows
• Benchmarks and evaluation protocols for reliable agentic behavior
• Applications in science, medicine, safety-critical systems, and decision support

Submission details

Submissions should present recent or ongoing work relevant to the workshop themes. The workshop is intended to provide a focused venue for discussion at the intersection of modern agentic systems and rigorous uncertainty quantification.

Submission portal

OpenReview submission site

Submission template

Submissions should use the official ICML template

Page limit

Submissions may be up to 8 pages of main content. Shorter submissions are also welcome.

Archival policy

Submissions are non-archival

Presentation

Accepted workshop papers will appear as posters; selected papers will be invited for contributed spotlight talks

Scope

Previously published, concurrently submitted, and ongoing work may be considered

* Please create your OpenReview profile at least two weeks before the submission deadline.