Workshop: FTXS: Workshop for Faults, Trustworthiness, and eXplainability for AI Systems at Scale
Authors: Scott Levy (Sandia National Laboratories), Bo Fang (The University of Texas at Arlington), and Hailong Jiang (Youngstown State University)
Abstract: FTXS convenes researchers and practitioners to address trustworthy AI systems at scale under real-world variability. Modern AI/LLM pipelines (training, inference, RAG, and agentic systems) face “gray failures” such as distribution drift, prompt/tool/RAG variability, infrastructure noise, and limited-precision effects. Small perturbations can amplify into hallucinations, instability, miscalibration, irreproducibility, and silent regressions — wasting GPU-hours and undermining confidence in AI-assisted decisions.
FTXS organizes discussion around a practical closed loop: Disturb → Degrade → Detect → Diagnose → Defend → Validate. We emphasize explainability as a diagnostic tool for attribution across components (model, retrieval, tools, infrastructure, precision), and we aim to synthesize shared taxonomies, measurement practices, and seed artifacts (checklists, benchmarks, and evaluation protocols) that improve both research and operational practice.
Website: https://sites.google.com/view/ftxs2026
Back to FTXS: Workshop for Faults, Trustworthiness, and eXplainability for AI Systems at Scale Archive Listing
Back to Full Workshop Archive Listing