B. Biggio

Several open research questions are identified, including how to efficiently record incidents and how to determine whether vulnerabilities and incidents generalize, and privacy requirements are summarized and research directions for the secure and trustworthy deployment of AI agents are outlined.

AI-based malware detectors that rely on domain knowledge significantly outperform deep learning models in long-term resilience and security against adversarial attacks.

Attack ensembles optimized for minimum-norm strategies can provide a more accurate and flexible evaluation of adversarial robustness than traditional fixed-budget methods.

Slash malware detection labeling costs by 90% using combined active and semi-supervised learning, without sacrificing performance.

Colluding federated learning participants can no longer hide: BlackCATT enables robust black-box traitor tracing even with sophisticated attacks and complex model architectures.