Abstract: Over the past two decades, we have witnessed the emergence of three transformative paradigms: cloud computing, edge computing, and artificial intelligence. Together, they have reshaped the computational landscape, enabling data and computation to seamlessly flow across enterprise systems, cloud platforms, edge infrastructure, and endpoint devices.

In this new era, data has become the new oil—a critical, high-value asset that is increasingly sensitive and in need of protection. While encryption of data at rest and in transit has become standard, a new frontier in security has emerged: confidential computing. This paradigm leverages hardware-based memory encryption and isolated execution environments (enclaves) to ensure both the confidentiality and integrity of applications, even in the presence of a potentially compromised system.

Confidential computing has already shown that many existing applications can run within these secure enclaves with minimal or no modification. As agentic systems—autonomous agents that span the full platform spectrum—gain traction, they will increasingly consume and process sensitive data without direct user oversight. This raises important concerns about the privacy of both models and user interactions, particularly in the context of large language models (LLMs) and other AI-driven services.

To address these concerns, we argue that confidential computing should serve as the foundation for agentic systems, providing the necessary trust guarantees. Furthermore, with the rise of AI workloads, AI accelerators must now become enclave-aware and enclave-compatible, ensuring that hardware acceleration does not compromise security guarantees.

In this talk, we explore the readiness of real-world applications for deployment on modern confidential AI platforms. We evaluate the performance overheads introduced by enclave execution, the compatibility challenges faced by AI frameworks and accelerators, and the overall feasibility of securing the next generation of intelligent, autonomous systems.

As organizations look to build secure, intelligent, and scalable systems, confidential computing offers a critical pathway to trust and compliance in a data-centric AI world.

Bio: Dr. Franke is a Distinguished Research Scientist at the IBM T.J.Watson Research Center, Yorktown Heights, NY since 1993. He  graduated with a Diplom Informatik degree (computer science) from the University of Karlsruhe (now KIT), Germany in 1987. He obtained a Ph.D. in Electrical Engineering in 1992 from Vanderbilt University, USA. His contributions to IBM’s Systems are wide ranging, covering the entire system stack from processor architecture, operating systems, performance optimization, compiler, HPC, middleware, cloud solutions and most recently confidential computing. He authored / co-authored 150 publications in peer reviewed conferences and journals in these domains and holds 181patents in these areas. He is an IBM Master Inventor and an ACM Fellow. Since 2011 he is an Adjunct Professor at the Courant Institute at New York University. In addition, since 2022 he is an Adjunct Research Professor, ECE, University of Illinois Urbana-Champaign and since 2024 an Adjunct Professor, EE/CS, Columbia University.