Trojan Code: Adversarial Machine Learning and Secure AI Systems

Kassem Kallas, Ph.D., HDR, EMBA, Senior Member IEEE, is a Senior Scientist at the French National Institute of Health and Medical Research (INSERM) and a Professor in collaboration with IMT Atlantique. He is internationally recognized for his contributions to Artificial Intelligence Security, Cybersecurity, and Adversarial Machine Learning, with impactful work spanning Europe, the United States, and the Middle East.Dr. Kallas earned his Ph.D. in Information Engineering and Science from the University of Siena (Italy), where he developed a game-theoretic framework for adversarial information fusion in distributed sensor networks. This foundational work influenced research in wireless sensor networks, cognitive radio systems, image forensics, and adversarial learning. In 2025, he completed the Habilitation à Diriger des Recherches (HDR) at the University of Western Brittany (UBO), the highest academic qualification in the French university system.His international career includes positions at prestigious institutions. From 2020 to 2022, he served as a Research Fellow at the U.S. National Institute of Standards and Technology (NIST), contributing to wireless communication and signal analysis research. He then joined Inria France, where he worked on the SAIDA project focused on securing AI systems for defense applications. He currently serves at INSERM, where he conducts research on the security and privacy of AI for healthcare applications and supervises doctoral work in these areas.Dr. Kallas contributes to several major French national research initiatives, including the PEPR “Secure, Safe, and Fair Machine Learning for Digital Health (SSF-ML-DH)”, and the CybAille Industrial Chair in Cybersecurity and Trustworthy AI for Health, which bring together partners such as IMT Atlantique, Université PSL, CEA-LIST, Thales, and CHU Brest.

• Chapter 1 Introduction.- Part I Foundations of Artificial Intelligence Security.- Chapter 2 Mapping the AI-Security Battlefield: Threats Across theMachine-Learning Lifecycle.- Chapter 3 Behind the Backdoors: Threats and Safeguards for Deep-Learning Systems.- Part II Backdoor Attacks and Defenses in Deep Neural Networks.- Chapter 4 Stealthy Clean-Label Backdoors: How an Image-Classification Model Can Be Attacked.- Chapter 5 Illumination-Modulated Video Backdoor Attacks on Anti-Spoofing Rebroadcast Detectors.- Chapter 6 Power Play: Backdooring DNNs Through Energy-Drain Triggers.- Chapter 7 Expecting the Next Move: Robust Backdoors under Non-IID Federated Training.- Chapter 8 When One Shield Is Not Enough: Layering Defenses Against Backdoor Attacks.- Chapter 9 Rare-Event Simulation for Black-Box Backdoor Defense.- Chapter 10 Game-Theoretic Modeling of BackdoorAttacker–Defender Dynamics.- Chapter 11 Cost-Constrained Backdoor Games in Deep Learning.- Part III DNN Watermarking for Intellectual Property Protection.- Chapter 12 Robust and Secure Watermarking for Deep Neural Networks.- Chapter 13 DNN Watermarking in Blackbox Settings using Image Mixup.- Chapter 14 Cryptographically Bound Mixup Watermarks for Black-Box DNNs.- Part IV Emerging Trends, Open Issues, and Future Research Directions in AISecurity.- Chapter 15 Security Horizons: Emerging Threats and Future Directions for Trustworthy AI.- Index.