Kyndryl's sovereign step-in plans: who runs your IT if access stops?

By DataForge team19 November 20265 min read

Kyndryl now sells sovereignty step-in planning, and NZ firms can borrow the method to test who'd run their cloud and AI systems if offshore access stopped.

What Kyndryl announced

On 12 November Kyndryl announced a new sovereignty service called Sovereign Step-in Readiness. Kyndryl says it'll work with each customer to define sovereignty scenarios, set roles and governance, write operating procedures and runbooks, and test whether the customer can take over defined operational responsibilities. Customers choose a readiness level. The lowest prepares people and processes, and the higher levels add localised tools and automation that give the customer more operational independence over data, automation and technology. The tools can run inside the customer's own systems or on a Kyndryl Bridge instance hosted in their region (Bridge is Kyndryl's management platform), which Kyndryl says uses integrated data and AI. The release says those regional instances are still being designed. Pricing, an availability date and a country list are all missing from the release.

IT Brief Australia covered the launch on 16 November. The article adds that Kyndryl provides advisory, implementation and managed services in more than 60 countries, and the article body doesn't mention Australia or New Zealand. Kyndryl supports the pitch with its 2026-27 Readiness Report, which it says shows under a quarter of leaders think their organisation could cope with geopolitical disruption. The survey method isn't set out in the release, so read the figure as a vendor finding.

Why the timing matters

On Friday 13 November an outage took Microsoft 365, Outlook and Teams offline for hours, and ASIO and the Australian Federal Police were among the organisations affected. Home Affairs Minister Tony Burke said the government's agreement with Microsoft "has in fact been activated today" so Australian services would be restored first. Microsoft told Burke it didn't think a cyber attack caused the disruption. The agreement is a memorandum of understanding signed on 9 June 2026. It covers data centre and network resilience, threat sharing and secure AI adoption in government, and it aims to keep vital services under Australian control.

Read together, the two events show where a typical NZ business stands. A national government can negotiate priority restoration with a hyperscaler, and a 200-person distributor in Penrose has no such leverage. The distributor's continuity rests on standard service terms and on whatever preparation it has done to work without the service. Kyndryl's service targets large enterprises with sizeable managed-service contracts. The method underneath it costs little to copy at a smaller scale: name the scenario, name the person who acts, write the runbook and rehearse it.

Where private AI fits

In NZ, sovereignty discussions usually stop at data residency, meaning whether the data sits in an Auckland region. Step-in planning goes further and asks whether your staff could keep a system running and keep using its data if the offshore service, its support team or its AI model became unavailable or legally restricted.

AI features make the step-in question harder to answer. Plenty of business software calls a model hosted overseas, and the company hosting the model is often different from the company selling the software. When the model goes down or gets withdrawn, the AI feature stops, even if your records sit in a local region. A workflow that now relies on AI document extraction, summarisation or agent actions needs a fallback path. For some workflows the fallback can be a model running on hardware you control. Kyndryl's higher readiness tiers, with localised tools and automation, point the same way. A small open-weight model on a local server won't match a frontier model. For invoice extraction or internal document search, a local model can keep work moving until the main service recovers. We set out how that kind of setup works on our private AI server page.

NZ privacy law gives firms room to use offshore processing and leaves the continuity question to them. The Privacy Commissioner's guidance on disclosing information outside New Zealand says IPP 12 doesn't apply where personal information goes to an agent for storage or processing and the agent doesn't use or disclose it for its own purposes. Offshore AI processing can therefore be lawful without extra contractual steps, provided the provider isn't reusing your data for its own ends. The guidance deals with disclosure only. Whether your business can keep operating while that provider is down is a decision for the board and the operations team.

What NZ firms should take from it

Most NZ organisations won't buy a sovereignty programme from a global integrator, and a full programme would be more than they need. A dependency map they can trust, plus a few rehearsed runbooks for the services that would hurt most to lose, covers most of the value. Expect the mapping to turn up a few surprises. Typical ones are admin accounts that exist only inside an overseas identity platform and AI features that nobody ever recorded as a dependency.

Kyndryl hasn't said when or where its regional Bridge instances will be available. If you're considering Kyndryl's service itself, ask for those dates and locations in writing. For everyone else, the planning method is the useful part, and it's available now without a vendor.

Practical next steps:

  • List every system with AI features. For each one, record where the model runs and who operates it, separately from where the data is stored.
  • Choose two scenarios to plan for: a multi-hour outage of your main productivity suite, and a week without access to one AI provider. Write down who makes the call, who does the work and what staff do in the meantime.
  • Keep a break-glass admin account and an offline copy of key contacts and procedures outside the platform you're planning for.
  • For each AI-dependent workflow, decide whether a manual fallback is enough or whether a locally hosted model should be kept on standby. Test the fallback at least once a quarter.
  • Ask your MSP and software vendors what step-in support they provide, what their contracts say about restoration priority, and how quickly they'll export your data in a usable format.
  • Add the runbooks to your incident and continuity processes so they get reviewed on the same cycle as everything else. Our ITIL 4 service management page covers how those processes fit together.

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
Get new articles by RSS