AI agents are enabling new opportunities for enterprises while raising demands for security, governance, and control. Deploying these agents and systems at scale requires a full-stack approach that extends governance and security beyond the model to the software and infrastructure layers where agents operate.
Today, Lenovo announced that it is working with NVIDIA to support the NVIDIA Open Agent Safety Platform, an open reference design for full-stack security and governance for AI agents across software and infrastructure layers. The NVIDIA Open Agent Safety Platform allows organizations to build stronger safeguards for agentic AI across cloud, hybrid, on-premises, and air-gapped environments. As part of our overall Lenovo Hybrid AI Advantage with NVIDIA portfolio, we will support this new open reference design to ensure full-stack governance and control through the software and infrastructure that runs AI agents.
The NVIDIA Open Agent Safety Platform uses a zero-trust architecture designed to put enforceable boundaries around how AI agents operate. Access is restricted by default, with permissions and controls designed to help organizations govern what agents can access and do across their environments. The design combines NVIDIA OpenShell, the secure runtime that governs agent execution, access and inference routing, with NVIDIA Sentry, an out-of-band watchdog running on NVIDIA BlueField-4 DPUs that monitors agent activity and independently enforces security policies in silicon quarantining agents in milliseconds.
Lenovo Hybrid AI Factory solutions will support the NVIDIA OpenShell secure runtime – running on NVIDIA Vera CPU and other processor platforms – and NVIDIA Sentry running on NVIDIA BlueField-4.
The collaboration between Lenovo and NVIDIA builds on a broader commitment to advancing AI security through openness and industry-wide cooperation. As we highlighted when we joined the Open Secure AI Alliance, securing AI at scale requires trust throughout the development and deployment lifecycle, along with collaboration across the technology ecosystem.
The NVIDIA Open Agent Safety Platform puts that approach into practice at the infrastructure level, giving organizations greater control over how agents interact with their systems, data, and applications while preserving flexibility across models and deployment environments. For Lenovo, this is an important part of advancing our vision of hybrid AI. We believe in giving customers choice in how and where they deploy AI while building security, governance, and trust throughout the stack.
That flexibility has to come with the right safeguards too. Security, governance, and customer control should be part of how AI is deployed from the start, particularly as agents take on more tasks and begin acting on behalf of people and organizations.
Looking ahead, putting these principles into practice at scale will take continued work across the industry. Broader enterprise adoption of agentic AI will depend on collaboration across the technology ecosystem. Lenovo will continue working with NVIDIA and our other partners to help make AI more secure, practical and responsible at scale.
Learn more about the NVIDIA Open Agent Safety Platform announcement here.