What Red Hat released
Red Hat announced OpenShift 5 on 9 November at KubeCon + CloudNativeCon North America in Salt Lake City. OpenShift 5 is the platform's first major version in seven years. Customers with existing subscriptions can get the release now, and OpenShift 4 clusters can move to it with an in-place update. Help Net Security puts general availability at 11 November, so different reports will show slightly different dates.
The sovereignty features are aimed at organisations that need tight control over where workloads run and who can reach them. Red Hat lists hardened multi-tenancy for sovereign deployments, post-quantum cryptography for control plane traffic, stronger network isolation, automated certificate lifecycle management and native support for disconnected deployments in air-gapped, edge and regulated environments. The release also adds automated GPU compute slicing, which shares accelerators between AI inference and conventional workloads. OpenShift 5 now runs virtual machines, containers and AI workloads under one control layer.
The other major change is in OpenShift Lightspeed, the platform's built-in assistant. Lightspeed has moved from answering questions in a chat window to acting as an agent that investigates operational problems, weighs likely causes and proposes fixes, and administrators decide whether each recommended action runs. One New Zealand's head of cloud and infrastructure platforms, Sanit Kumar, said the features suit the telco's 5G core and enterprise workloads.
Why an admin agent raises a data sovereignty question
Lightspeed answers a request by sending the administrator's question to a large language model, along with a system prompt and, where the context window allows, retrieved documentation and conversation history. When the agent analyses logs, extracts from those logs go into the prompt as well. Cluster logs routinely hold usernames, internal hostnames, IP addresses and API paths, and sometimes customer identifiers that an application wrote by mistake. A sovereign, disconnected cluster loses much of its value if its troubleshooting agent sends log extracts to a model hosted in another country.
The customer chooses the model. The open-source Lightspeed service lists OpenAI, Azure OpenAI, IBM watsonx, OpenShift AI and RHEL AI as officially supported backends. Its configuration includes vLLM provider types for models you host yourself on OpenShift AI or RHEL AI, and the project also mentions local options such as an Ollama server. The service has a redactor that strips content from questions using regex filters defined in its configuration file. The project page doesn't publish default patterns, so we'd assume the redactor removes nothing until your team writes and tests its own rules.
Red Hat hasn't said publicly which backend it recommends for the new agent mode, or whether agent mode behaves the same on a small self-hosted model as on a large hosted one. Neither the launch coverage nor the project page answers those questions. Teams planning a fully local deployment should get answers from Red Hat before they rely on a self-hosted model for agent tasks.
What OpenShift 5 means for NZ organisations
Few mid-sized NZ businesses run OpenShift directly, but the pattern behind the release applies much more widely. Infrastructure vendors are building AI agents into admin consoles, monitoring tools, backup products and service desks, and each agent needs a model hosted somewhere. Most data residency policies we see were written for business records in Microsoft 365, an ERP or a CRM. Those policies rarely mention operational telemetry, even though logs, alerts and configuration dumps are exactly what an ops agent consumes.
For organisations that do run OpenShift, such as banks, telcos, government agencies and their suppliers, version 5 makes a fully local AI stack more practical. GPU slicing lets a modest on-premises GPU footprint serve a self-hosted model for Lightspeed alongside other inference work. Native disconnected support lets the whole stack run without outbound calls. We see the sovereign multi-tenancy features as most useful to NZ managed service providers that host several clients on shared clusters and have to prove tenant separation to auditors.
The Privacy Act angle is easy to miss because ops data doesn't feel like customer data. When personal information in your logs reaches a model provider outside New Zealand, your privacy assessment has to account for that flow. Information Privacy Principle 12 sets the conditions for disclosing personal information overseas, including a reasonable belief that the recipient protects it to a comparable standard, for example under contract clauses. Agreeing contract terms with a provider takes longer than switching on an assistant, so legal and procurement checks should start before the upgrade window.
Practical next steps
- List the AI features already in your infrastructure tools. Cover monitoring, backup, endpoint management, ITSM and container platforms, and record which model each feature calls and where that model is hosted.
- Choose the Lightspeed backend before you upgrade. If you run OpenShift and want logs to stay onshore, plan for a self-hosted model on OpenShift AI or RHEL AI and size the GPU capacity with slicing in mind. Our local AI server page covers the hardware and hosting questions that come up.
- Write and test redaction rules against real log samples. Pull a week of logs, check what the regex filters catch and what they miss, and repeat the test whenever applications change their logging.
- Keep human approval on every remediation action. Treat agent-proposed fixes as standard changes in your change process, with the approver and outcome recorded.
- Ask Red Hat in writing which backends support agent mode and what usage data, if any, leaves the cluster.
- Add operational data to your data residency policy. Name logs, alerts, traces and configuration exports, and set a rule for which AI services may receive them. If you want help setting those rules, our AI consultancy team works on AI governance for NZ and Australian organisations.
Sources
- Red Hat launches OpenShift 5 with AI troubleshooting (SecurityBrief NZ)
- Red Hat OpenShift 5 brings autonomous troubleshooting to hybrid clouds (Help Net Security)
- Red Hat Launches OpenShift 5 With Agentic AI Troubleshooting (Let's Data Science)
- openshift/lightspeed-service (GitHub)
- Principle 12: Disclosure of personal information outside New Zealand (Office of the Privacy Commissioner)
Want to know what this means for your organisation? Our team in Auckland can show you your top three opportunities in a free 30 minute assessment.
Get my free assessment How we help