AIHWS

The evolution of machine learning (ML) as an enabling technology has ushered in a new era of possibilities, including its integration into security-critical applications. AI-driven techniques are increasingly being leveraged to enhance threat detection, automate vulnerability assessments, and strengthen defense mechanisms in complex and dynamic systems. At the same time, federated learning (FL) has emerged as a pivotal advancement in distributed intelligence, enabling privacy-preserving collaboration across sensitive domains such as healthcare, finance, and autonomous systems. However, alongside these advancements, both FL and broader ML applications introduce a range of new security challenges. Poisoning attacks, adversarial manipulations, and inference threats undermine the integrity, confidentiality, and trustworthiness of ML models. Addressing these risks demands a holistic approach that incorporates security-by-design principles, the use of AI for proactive defense, and a thorough understanding of vulnerabilities inherent to ML-enhanced security mechanisms.
In this talk, we explore the full lifecycle of ML model development and deployment through a security lens, analyzing attack vectors and defense strategies at each stage. We highlight emerging threats, present current research efforts in building adversarially resilient systems, and outline key challenges in developing trustworthy, robust, and attack-resilient ML frameworks.

Over the past six years, Google has been working with the open source community to build OpenTitan, the first open source silicon Root of Trust chip. This silicon will be the first broadly used Root of Trust chip at Google with a fully transparent design. This talk will review key elements that make OpenTitan a success: The open source collaboration, the progressive hardening, the PQC support, and highlight contributions from research.