Dell outlines AI zero-trust controls

Dell Technologies used its Dell Technologies World event in Las Vegas on May 18 to present a security architecture for agentic AI that centered on local deployment, confidential computing and sandboxed runtime controls. The company said its new Dell Deskside Agentic AI offering lets enterprises run agentic workflows locally to address cost, latency and data sovereignty constraints, while support for NVIDIA OpenShell adds controls for building, testing and governing AI agents. Dell’s framing puts security controls around the infrastructure that runs autonomous software, rather than relying only on model-level safeguards. NVIDIA describes OpenShell as an open-source runtime for executing autonomous AI agents in sandboxed environments with kernel-level isolation and policy controls that can restrict access to files, credentials and external networks. ### Which Dell announcement actually introduced the new controls? (dell.com) Dell Technologies said on May 18 that Dell Deskside Agentic AI became a new part of the Dell AI Factory with NVIDIA, aimed at workgroups that want to deploy and scale agentic AI workflows locally. The company said the system is powered by Dell workstations, the NVIDIA NemoClaw software stack and Dell services, and is designed to handle models ranging from 30 billion to 1 trillion parameters. (docs.nvidia.com) A second Dell release published the same day said OpenShell integration extends across the Dell AI Factory, from deskside systems to Dell PowerEdge XE servers in the data center. Dell said that setup is intended to give enterprises a local and “secure” foundation for production-ready agentic AI on infrastructure they control. (dell.com) ### How is Dell describing the risk around autonomous agents? NVIDIA said in a March 23 technical post that OpenShell is meant to add security in the infrastructure policy layer, rather than in the model or application layer. The company said autonomous agents create new risks because they can act on tools and data stores, making containment and policy enforcement part of the runtime environment. (dell.com) NVIDIA’s OpenShell documentation says teams can use declarative policies and sandbox runtime controls so agents do not receive unrestricted access to local files, credentials or outside networks. That maps to the types of enterprise concerns Dell highlighted around governance, privacy and controlled execution for agentic systems. ### Where do data localization and confidential computing fit? Dell said its deskside offering is aimed at customers that want to avoid the “data sovereignty constraints of cloud-only approaches,” and another May 18 release said enterprises can keep data inside their own environment when running secure autonomous agents locally. (blogs.nvidia.com) Those statements tie Dell’s security pitch to on-premises or customer-controlled deployment rather than public-cloud dependence. (docs.nvidia.com) NVIDIA said in a March 23 blog on “confidential AI factories” that zero-trust AI environments require hardware-enforced trusted execution environments and cryptographic attestation. NVIDIA’s AI enterprise documentation says confidential containers are designed to isolate data and code from the host operating system, hypervisor and privileged infrastructure operators during execution. (dell.com) ### What does “zero trust for AI infrastructure” mean in this context? NVIDIA said zero-trust AI factories are built to remove implicit trust in the underlying infrastructure by using trusted execution environments and attestation. In practice, that means verifying the environment running the workload and limiting what operators, hosts or connected services can see or access. (developer.nvidia.com) Dell’s May 18 materials did not publish a separate zero-trust framework document in the sources reviewed, but the controls it named were specific: local deployment, governance for agentic workflows, OpenShell sandboxing and infrastructure that scales from workstation to server. Those pieces show how Dell and NVIDIA are packaging identity, isolation and segmentation controls around AI workloads at Dell Technologies World. (developer.nvidia.com) ### What comes next from Dell and NVIDIA? Dell Technologies World ran in Las Vegas from May 18 to May 21, according to Dell and NVIDIA event pages, and Dell’s 2026 press kit collects the company’s AI and infrastructure announcements from the conference. NVIDIA’s OpenShell documentation and technical blogs provide the implementation detail behind the sandboxing and policy controls Dell referenced in its product launch materials. (dell.com 1) (dell.com 2)