NVIDIA published an article on its official blog, emphasizing that AI safety is fundamentally an engineering problem that needs to be addressed by defining safety requirements, implementing enforceable controls, clarifying ownership, and verifying safeguards. The article points out that as AI capabilities grow, the industry must accelerate safety engineering practices, broaden the accessibility of defensive tools, and share effective solutions more quickly. NVIDIA also mentioned OpenShell as an open-source security runtime that can enforce policies within the agent stack and provide a sandboxed execution environment, with partners such as Cisco and JFrog already building related security solutions based on OpenShell.