Security in High-Performance Computing (HPC) has traditionally been an "operational" challenge (i.e., restrict access and usage to certified users). However, as HPC gradually permeates more areas of public interest, a hands-off approach to security aspects in favor of performance and power is becoming imprudent at best. Paired with HPC's traditional role of early technology adoption, a new set of early target-worthwhile vulnerabilities are emerging that are not necessarily found in other computing scenarios that operate with more established technologies.
In addition to early adoption vulnerabilities, the exploits specific to the HPC community arise from acute hardware heterogeneity (including distributed computing ecosystems); novel network technologies/topologies; massive resource management orchestration including power, compute, networking, and storage; heavy reliance on open software by volunteer overworked programmers; brittle experimental software not hardened by numerous deployments; and dusty deck software with no regular maintenance schedule. Further complicating this landscape is the integration of numerical HPC with data-driven methodologies, including Machine Learning, expanding potential attack surfaces to include data capture, storage, and transmission. In combination with commercial, single-node exploits, these vulnerabilities open fertile and complex new attack surfaces.
This workshop focuses on threats and solutions across the HPC hardware/software stack. These threats include weaknesses in current and future architectural designs, escalation of privileges through data extraction or computation manipulation, and intentional misuse of resources across scientific instruments feeding HPC machines.
We solicit papers on all topics related to security of high-performance computing, from hardware to run-time software to applications. We welcome both papers exploring new attacks and threat models as well as defenses. The topics of interest include, but are not limited to, the following:
As organizations adopt AI agents that execute code, access data, and coordinate across shared HPC infrastructure, every allowed action becomes a security decision. This talk explores how NVIDIA OpenShell brings isolation, deterministic policy enforcement, and controlled network access, unlocking actual enforceable agent runtime boundaries. Safe adoption requires extending zero trust to agent behavior, making continuous monitoring and governance foundational to how organizations grant, oversee, and govern agent autonomy.
Tim Klawa is a Public Sector Field Engineering Leader at NVIDIA, where he helps federal customers and partners design and implement accelerated AI solutions with a focus on cybersecurity. Prior to joining NVIDIA, Tim served as Public Sector CTO at Groq. Throughout his career, he has focused on architecting and optimizing AI systems that transform complex data into actionable intelligence. His experience spans building nation-scale data management platforms, optimizing AI-driven analytics, to orchestrating agent swarms in a safe and governed way.