From the Red Hat edition of October 29, 2026
Red Hat AI Factory with NVIDIA accelerates agentic AI security capabilities across the open hybrid cloud
Red Hat (with NVIDIA) · Strategic partner platform: agent runtime security and governed inference · Global (vendor announcement; no SLED deployment named)
- Publisher
- Red Hat newsroom
- Original publication
- Red Hat press release, Oct. 21, 2026
- Source retrieved
- 2026-10-30
- Event date
- 2026-10-21
What happened
Vendor claim: Red Hat says it is working with NVIDIA on the OpenShell Reference Architecture, an open-source agent sandbox runtime, and is actively collaborating to integrate it into Red Hat AI and Red Hat AI Factory with NVIDIA as a native platform capability. The release also describes a joint inference reference design with Model-as-a-Service access control, an AI gateway, vLLM and llm-d, plus self-service model discovery, rate limiting and entitlement tracking. The integration is described as in progress. The release gives no general-availability date, customer deployments or performance measurements.
Why it matters
Relevant to state CIO offices, university research IT and large county IT shops that are being asked to host AI agents on-premises or in a sovereign hybrid cloud. Per-tenant model access control, rate limiting and showback match how shared-service IT organizations charge back agencies or departments. No SLED customer is named, so applicability is inferred.
Evidence and measured results
Read the full press release on redhat.com. It is a forward-looking vendor announcement with a forward-looking-statements disclaimer. No benchmark, test method, sample or customer outcome is reported. The NVIDIA hardware claim that Vera CPUs deliver 2X faster agentic CPU performance comes from NVIDIA and gives no baseline.
Limitations and uncertainty
Vendor announcement only. The OpenShell integration is not yet a shipping native capability, no SLED deployment is cited, and performance claims have no independent validation.
Put this evidence to work
Lighthouse Advisory interpretation, grounded in this source. Enriched 2026-10-30; this does not change the original publication date. Each role follows the same checklist so takeaways compare across sources.
Sales
Role takeaway
- Customer problem
- Central IT is asked to host AI agents for agencies or departments but lacks a governed way to stop agents from reaching systems and credentials they should not touch.
- Stakeholders
- State or university CIO/CTO
- CISO
- Platform engineering lead
- Agency or departmental AI program owners
- Discovery questions
- Which agent use cases are being requested, and which systems would they touch?
- Do you already run OpenShift and OpenShift AI, and on which version?
- How do you currently meter and charge back shared AI model usage?
- What approval is required before an agent can write to a system of record?
- Value hypothesis
- A single platform layer for agent sandboxing and model access control could reduce one-off security reviews for each agent pilot, if the integration ships as described.
- Potential engagement
- Agent-hosting architecture workshop plus a sandboxed, non-production agent pilot on existing OpenShift AI.
- Unsupported claims
- That OpenShell is a shipping, supported Red Hat AI feature today
- Any performance or cost savings from the NVIDIA hardware claims
- That the architecture satisfies a specific state security standard or CJIS/HIPAA requirement
Pre-sales engineering
Role takeaway
- Fit
- Fits customers with OpenShift-based AI serving who want on-premises or hybrid agent execution with external policy enforcement.
- Architecture and integration
- OpenShell sandbox pods on OpenShift for agent execution; AI gateway and MaaS for model access and quotas; vLLM/llm-d for inference; NVIDIA GPU operator underneath.
- Prerequisites
- Current OpenShift AI release with MaaS/gateway components
- CNI that enforces Kubernetes NetworkPolicy
- GPU capacity and NVIDIA operator
- Identity integration for per-tenant entitlements
- Constraints
- OpenShell integration described as in progress, not native today
- No published performance data
- Hardware claims tied to upcoming NVIDIA platforms
- Security
- Enforcement outside the agent is the intended model. Verify sandbox defaults, egress enforcement and credential handling independently, because the release offers no test evidence.
- Proof of value
- Demonstrate that a sandboxed agent cannot reach an unapproved endpoint on the customer's cluster
- Demonstrate per-tenant rate limits and usage showback through the gateway
- Confirm which components are GA versus preview on the customer's subscription
Delivery
Role takeaway
- Work
- Stand up a sandboxed agent pilot environment, gateway entitlements and model serving, with documented policies and change control.
- Dependencies
- OpenShift AI version supporting the gateway features
- Red Hat roadmap delivery of OpenShell integration
- Security team policy review
- Ownership
- Central platform engineering team, with the CISO office owning sandbox and egress policy approval.
- Skills and adoption
- Train platform engineers on agent sandbox policy authoring, gateway entitlement management and vLLM serving operations.
- Governance checkpoints
- Security review of sandbox policy before first agent run
- Approval gate before an agent gets write access to any system
- Quarterly review of model entitlements and usage
- Acceptance criteria
- Proposed: 100% of egress attempts to non-allowlisted hosts blocked in pilot tests
- Proposed: per-tenant usage reports produced monthly
- Proposed: no production credentials present in the agent environment
- Risks
- Roadmap slip leaves the pilot dependent on unsupported components
- Permissive default policies left unchanged
- Agent scope creep before governance is in place
Implementation considerations
Lighthouse Advisory interpretation across the operating dimensions a public-sector buyer must settle before this evidence becomes a design. Each note answers the question under its heading for this specific source.
Architecture and integration
What must connect, and where does the AI sit in the workflow?
Agent execution would move into a sandbox runtime enforced outside the agent process, sitting beside OpenShift AI model serving (vLLM/llm-d) and an AI gateway for model access. Agencies planning agent pilots on OpenShift should keep the sandbox, the gateway and model serving as separate layers.
Governance
Who approves, reviews and stays accountable for outcomes?
Centralized MaaS access control and entitlement tracking could support agency-level usage policies and chargeback. Because the capability is not yet native, governance cannot rely on it today.
Security and privacy
What data, permissions and controls need testing?
The stated principle is that agent controls belong in the runtime and infrastructure, not inside the agent. An independent source review in this edition finds the OpenShell enforcement design sound but several defaults permissive.
Accessibility and workforce
Who is affected, and what skills or accommodations follow?
Platform teams would need Kubernetes, policy-as-code and agent-sandbox skills. The release does not address end-user accessibility.
Procurement
What should contracts, pricing and exit terms secure?
Components are offered through partners including Cisco, Dell Technologies and Lenovo, so state contract vehicles for those OEMs may be the purchase path. Buyers should separate shipping features from roadmap items in solicitations.
Operating model
Which teams own the service once it runs?
Positions Red Hat AI as a shared internal AI service with per-tenant metering, which suits central IT acting as a service provider to agencies or colleges.
Publication history
- 2026-10-29Red Hat · Issue 018 resources
Stable resource ID: redhat-ai-factory-nvidia-openshell-agent-security-reference-architecture-20261021