Program

Workshop programme · Afternoon session

Explaining, Auditing and Overseeing AI Systems

Time
October 6, 2026 - 14:00–16:00
Room
TBA
Duration
120 min
Talks
7
Talk slot
12 min + 3 min Q&A
Time budget

–10 min

Opening

Welcome and overview of the session

–45 min · 3 talks

Part I

Explaining LLMs, Explaining with LLMs

  1. –

    Explaining Bias-Refusal Geometry Across Languages: An Activation-Steering Study of Llama 2 and GPT-OSS

    Mehuly Chakraborthy and Antonio Rodà

  2. –

    Explainable Retrieval-Augmented Generation for Multilingual Dark-Pattern Auditing and Digital Consumer Protection

    Jobin Roy, Akshita Sakshi, Shifin Mohammed, Het Darshan Mehta, Marco Polignano, Giovanni Semeraro and Ernesto William De Luca

  3. –

    LLMs as Verbal Translation Layer for TSC Explanations

    Felix Marti-Perez, Cesar Ferri, Jan Arne Telle and Christian Bontveit

–30 min · 2 talks

Part II

Auditing Visual Perception Models

  1. –

    OCR-Guided Occlusion Sensitivity for Auditing Document Classifiers

    Valerio Caravani, Riccardo De Cesaris and Paolo Merialdo

  2. –

    TRACE-Pave: A Neural Perception and Auditable Decision Framework for Explainable Road Crack Assessment

    Tayyab Rehman and Giovanni De Gasperis

–30 min · 2 talks

Part III

Assurance and Human Oversight

  1. –

    From Post-Hoc Explanations to Assurance-Oriented XAI in Safety-Critical Automatic Target Recognition

    Vanessa Buhrmester, David Münch and Michael Arens

  2. –

    Following Delegation Traces to Study Human Oversight in Interaction with Multimodal AI

    Agnese Chiatti, Matteo Matteucci and Viola Schiaffonati

–5 min

Closing remarks

Wrap-up of the session