Joint Workshop

Security, Human Awareness, and Risk Mitigation in AI-driven Systems

SHIELD-AI 2026 is an interdisciplinary workshop on AI risk management, human–AI interaction, and human-centered cybersecurity in adversarial and security-critical environments.

Held within ECML PKDD 2026
7–11 September 2026 · Naples, Italy

11 September 2026

Workshop Program

10:30–11:15

Keynote

Building 1, Via Claudio, Room II3

Beyond the Network: Understanding Cognitive Threats

Daniel Díaz-López, University of Murcia

11:15–12:30

Paper Presentations · Session I

Session Chair: Pantaleone Nespoli
Building 1, Via Claudio, Room II3

SOFIA: A Unified Framework for Deepfake-Driven Terrorism and Hybrid Threats

Vita Santa Barletta, Danilo Caivano, Gabriel Cellammare, Pietro Chiedi, Antonio Piccinno

Consensus-Driven Countermeasures: A Multi-Agent System Leveraging Open-Source Large Language Models to Mitigate Disinformation Threats

Sebastian Kula, Martin Tamajka

Interaction Cyber-Ranges for Human–AI Companion Systems: Stress-Testing Trust-Relevant Behaviour under Degraded Semantic States

Roberta Presta, Lorenzo Bacchiani, Andrea Bedei, Mario Bravetti, Andrea Melis, Giovanni Pau, Roberto Girau

Improving Fairness in Deepfake Detection via Redundancy-Based Undersampling

Yan Cunha, Daniel Moraes, Gabriel Prenassi, Guilherme Fonseca, Julio Cesar Duarte, Leonardo Rocha, Sérgio Colcher

RedTreez – A Dynamic Tree-based Multi-turn Red-Teaming Approach

Iago Paulo, João Magalhães, David Semedo

A Digital Twin to Improve Cyber-Social Impact

Vita Santa Barletta, Danilo Caivano, Gabriel Cellammare, Andrea Dimauro, Pasquale Mesagna, Massimiliano Morga

12:30–13:30

Paper Presentations · Session II

Session Chair: Stefano Cirillo
Building 1, Via Claudio, Room II3

A Zero-Shot Evaluation of Large Language Models for Detecting Algorithmically Generated Domains

Donato D'Ambrosio, Francesco Cerasuolo, Alfredo Nascita, Antonio Montieri, Antonio Pescapè

Trend-based Multi-Modal Anomaly and Cyber Threat Detection with Traceable Explainability for Cloud Services

Mert Nakip, Rafał Gibała, Anna Grygar, Sławomir Nowak

When the Analyst Scans for the Attacker: Subverting Agentic LLM Security Operations in OT Networks

Jaafer Rahmani, Eren Çil, Axel Sikora

Security-Aware RF Manifold Learning: Anomaly Detection in 6G Networks via Contrastive CSI Representations

Andrea Piroddi, Andrea Melis, Alessandro Martini, Roberto Girau

Preventing Data Exfiltration at the Enterprise AI Boundary: A Progressive Guardrail Architecture

Thomas Schuster, Marian Lambert, Nico Doering

13:30–14:30

Lunch Break

Building 1, Via Claudio, Level -1

14:30–16:00

Paper Presentations · Session III

Session Chair: Roberta Presta
Building 1, Via Claudio, Room II3

Human-Centred Risk Mitigation for AI-Mediated Information Manipulation: A SOCMINT Framework Based on Information Manipulation Sets

Antonio Scala

From Manipulation to Security-Relevant Action: Human Exposure in AI-Mediated Cyber Risk

Luigi Di Biasi, Francesca Formisano, Giuseppe Genito, Francesco De Micco, Gennaro Sepe, Lorenzo Di Palo, Serena Lembo

Should LLM Agents Decide in Social Simulations? Comparing Finite-State and LLM-Based Decision Policies

Alejandro Buitrago López, Javier Pastor-Galindo, José Antonio Ruipérez-Valiente

The FRIAct Framework: Addressing the Methodological Gap Concerning Fundamental Rights Impact Assessment

Eylül Erva Akin, Nicole Inverardi, Ilaria Penco, Daniele Scelta, Silvia Tessaro Trapani, Manuela Zaidan

Towards a Methodology for Calibrating Human-in-the-Loop Trust in Clinical AI Using Standardized Reason Codes and Adaptive Breach Thresholds

Luana Alvarenga, André Moreira, Caio Justino, Igor Fernandes, Luana Oliveira, Anna Araújo, Marcos Farias, Bruno Nunes, Leandro Rodrigues, André Peres, Renan Moioli, César Rennó-Costa

Risk-Aware Semantic Grounding for Trustworthy LLM-Based Robot Planning

Łukasz Sobczak, Nur Keleşoğlu, Sławomir Nowak

GAIMP: Clinical Text Sanitization via Agent-Based Generative Anonymization

Maurizio Atzori, Stefano Cirillo, Alessandro Pani, Giuseppe Polese, Giandomenico Solimando

16:00–16:30

Coffee Break

Building 1, Via Claudio, Level -1

16:30–18:10

Paper Presentations · Session IV

Session Chair: Eliana Pastor
Building 1, Via Claudio, Room II3

Inspecting Training Dynamics of Similarity Development in Supervised Vision Networks

Katarzyna Filus, Mateusz Żarski

When In-Distribution Is Not In-Scope: Empirical Evidence for Out-of-Model-Scope Failures

Raul Ferreira, Leandor Alvim

Agentic Adversarial Debiasing: A Dynamic Multi-Agent Framework for Fairness-Aware Machine Learning

Wassim Fathallah, Nahla Ben Amor

Ghost Echoes: Semantic Erasure Failure in Retrieval-Backed Applications

Allan Mukkuzhi, Jacob Kammerzell, Noella Uwayisenga, Yeabsira Bizualem, Chandranil Chakraborttii

Self-hosted LLMs for Phishing Detection: An Empirical Evaluation of Accuracy and Efficiency

Aleksandra Rząca, Wojciech Matuszyński, Kamil Faber, Łukasz Faber

A Novel AI-Assisted Knowledge Discovery Service to Support Drug-Manufacturing Hypotheses

Federico Negri, Vito Morreale, Ernesto La Mattina, Marialuna De Tommaso, Valentina Mazzonello

Mind the Gap: Robustness Risks in PII Detection Systems

Adeel Zafar, Sławomir Nowaczyk

Affective Asymmetry in News Reactions across Reddit and Bluesky

Mario Trerotola, Anna Esposito, Mimmo Parente

18:10–18:40

Round Table

Building 1, Via Claudio, Room II3

18:40–19:00

Conclusion & Remarks

Building 1, Via Claudio, Room II3

About the workshop

Artificial Intelligence systems are increasingly integrated into everyday applications, digital infrastructures, and security-critical environments. While these systems offer significant benefits, they also introduce systemic risks related to technical vulnerabilities, algorithmic bias, integration issues in complex systems, and the dynamics of human–AI interaction.

The workshop is held within ECML PKDD 2026 — the European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases — taking place in Naples, Italy, from 7 to 11 September 2026.

At the same time, modern cyber threats have shifted toward the cognitive domain, exploiting human behavior, social dynamics, and large-scale information ecosystems. Social cyber attacks such as phishing, social engineering, and disinformation campaigns manipulate trust mechanisms, cognitive biases, and online communication patterns.

SHIELD-AI 2026 provides an interdisciplinary forum that brings together researchers and practitioners from machine learning, knowledge discovery, cybersecurity, and human-centered computing. The workshop aims to advance methods for identifying, modeling, and mitigating risks in AI-enabled socio-technical systems operating in adversarial and human-critical environments.

Important dates

~~5 June 2026~~

12 June 2026

Paper Submission

~~1 July 2026~~

8 July 2026

Paper Acceptance

Full-day event

Workshop format

Topics of interest

Risk identification and modeling in AI-driven systems and multi-component ecosystems.
Human–AI interaction risks, trust, overreliance, and decision-making under automation.
Human-centered cybersecurity and social cyber attacks, including phishing, social engineering, and disinformation.
Machine learning and knowledge discovery for cyber threat detection on large-scale heterogeneous data.
Natural language processing and large language models for malicious or manipulative content analysis.
Graph learning and knowledge graphs for social networks, communication patterns, and cyber threat ecosystems.
Bias, robustness, uncertainty, and adversarial machine learning in human-critical settings.
Explainable and trustworthy AI for security-critical applications.
Generative AI for both cyber attacks and defensive analysis.
Runtime monitoring, continual risk assessment, and adaptive mitigation strategies.
Human behavioral and psychophysiological data analysis.
Privacy, data protection, cyber threat intelligence, and knowledge sharing.
Human-in-the-loop and reasoning-based methods for dynamic risk assessment and explanation.
Case studies and empirical evaluations in real-world deployments.

Submission guidelines

Papers must be written in English and formatted in LaTeX following the Springer LNCS author kit. The maximum length is 16 pages including references. Over-length papers will be rejected without review.

Up to 10 MB of additional material may be uploaded with the submission. Appendices should be submitted separately, and the main paper must remain self-contained within the page limit.

SHIELD-AI 2026 welcomes regular research papers, survey articles, mature work, recent research contributions, preliminary exploratory work, and industry experience or case-study papers. Submissions must be original and not under review elsewhere.

All submissions should be made in PDF via Microsoft CMT and must adhere to the Springer LNCS style. At least one author of each accepted paper must register and present the paper in person in Naples.

The workshop papers may be included in a joint post-workshop proceeding published by Springer Communications in Computer and Information Science (CCIS), with authors able to opt in or out.

How to submit

Open the submission link here
Click on "Create New Submission" and select SHIELD-AI 2026 from the list of workshops.
Fill in the required submission information and upload your manuscript.