Research papers · 2025.12
Taming Silent Failures: A Framework for Verifiable AI Reliability
Taming Silent Failures: A Framework for Verifiable AI Reliability
Overview
FAME combines offline formal synthesis with online monitoring to detect confident but incorrect AI outputs. The authors demonstrate an autonomous-vehicle perception case and discuss the framework in relation to ISO 26262 and ISO/PAS 8800.
Author-supplied source information
Author-supplied information on arXiv.
IEEE Reliability Magazine ( Volume: 2, Issue: 4, December 2025)
This preprint has been accepted by IEEE Reliability Magazine. 10 pages, 3 figures
Version: arXiv v1 — manuscript revision