AI Agent Security Risks, CUDA Portability, and Kubernetes Adoption at Scale | TSG Ep. 1006
Alan Shimel, Mike Vizard, Chris Blask, Kate Scarcella, and Sid Nag examine security flaws in Model Context Protocol (MCP) servers and clients—an emerging layer of infrastructure that enables AI agents to access enterprise data and execute actions.
The panel discusses why MCP-related vulnerabilities matter as agentic architectures expand, and what security teams should prioritize when granting AI systems access to sensitive resources and workflows.
Next, the gang looks at an effort to improve portability for GPU-accelerated workloads by translating NVIDIA CUDA to AMD ROCm, and what that could mean for infrastructure flexibility and vendor lock-in.
The episode closes with a look at cloud-native maturity and the state of Kubernetes, following a Cloud Native Computing Foundation survey reporting 82% Kubernetes adoption—plus what “adoption” really means across production environments.
