What does Lenovo's new security platform mean for AI agents in your infrastructure?

The platform sits between agents and the systems they touch

AI agents now schedule workloads, move data, and call APIs without a human in the loop. That autonomy creates a new attack surface. The NVIDIA Open Agent Safety Platform addresses it with a zero-trust architecture that restricts access by default. Permissions and controls govern what agents can reach and do across cloud, hybrid, on-premises, and air-gapped environments. Lenovo will integrate this reference design into its Hybrid AI Factory solutions.

Two components enforce policy in different places

NVIDIA OpenShell runs as the secure runtime. It governs agent execution, access, and inference routing on NVIDIA Vera CPU and other processor platforms. NVIDIA Sentry operates out of band on NVIDIA BlueField-4 DPUs. The watchdog monitors agent activity and independently enforces security policies in silicon. It can quarantine agents in milliseconds.

Agents move fast.

The design preserves model and deployment flexibility

Organizations keep choice in how and where they deploy AI. The platform works across models and environments without locking customers into a single stack. Security, governance, and customer control become part of deployment from the start. Lenovo frames this as essential to its hybrid AI vision. The company joined the Open Secure AI Alliance earlier to advance the same principle at the development lifecycle level.

What the announcement does not cover

Pricing for the integrated solutions is not disclosed. A timeline for general availability across all processor platforms is not provided. The announcement does not name specific enterprise customers or pilot programs. Benchmarks comparing quarantine latency against software-only approaches are absent. How the platform interacts with existing identity and access management systems is not detailed.