NIS2 is not „another GDPR to tick off." For manufacturers caught by the national rules implementing it, it means concrete obligations around risk management, supply-chain security, and incident reporting. And since more and more document work is moving into AI tools, the question „will our AI vendor survive an audit" stops being theoretical. This piece pulls together, in one place, what actually applies and how to prepare, without legal jargon and without scaremongering. It is not legal advice, just a practical way to organize the topic from an AI-deployment angle.
What changed, and why now
The new rules implementing NIS2 took effect in spring 2026. For many manufacturers, that means they formally land in a regulated category for the first time. Registration and first-audit dates are staggered, but preparation starts much earlier, because it touches processes and suppliers you cannot swap overnight. Anyone who leaves this to the last quarter before an audit will be making technology decisions under pressure, which is the worst possible moment to pick an AI vendor.
Who this applies to
The regulation targets organizations important to the functioning of the economy and infrastructure. In practice, many mid-sized and large manufacturers fall into scope as a key entity under NIS2, often without realizing it. If you are not sure whether you are covered, that uncertainty is itself a signal to settle the question formally rather than assume „this doesn't apply to us."
The obligations that hit AI hardest
From a deployment angle, three areas of the regulation matter most.
- Risk management. You need to know where your data is, who does what with it, and be able to demonstrate it. An AI tool that ships your documents to a public cloud with no control over where they end up makes that hard.
- Supply-chain security. Your suppliers become part of your risk surface. An AI vendor is a supplier like any other: you must be able to answer where it processes data, on what infrastructure, and what happens when something goes wrong.
- Incident reporting. You need a record of what happened. An AI system that leaves no auditable trail (who asked, which sources an answer is grounded in) is a problem here, not a help.
Why public-cloud AI becomes an audit problem
Most popular AI tools are public-cloud services, shared across many customers, processing data outside your perimeter. For a team that has to demonstrate control over its data and its supply chain, that is a hard starting position. The point is not that public AI is „bad," but that by default it does not answer the questions an auditor will ask. We wrote about this at length in the context of why public AI won't pass your audit.
How to prepare, step by step
- Formally establish whether you are in scope, and document it.
- Inventory where your sensitive documents go today, including AI tools used informally by teams.
- Treat AI vendors like any other supplier in your supply-chain risk assessment.
- Require a clear answer from your AI vendor on where data is processed and whether an auditable trail remains.
- Do not leave the technology choice to the quarter before an audit.
What to ask your AI vendor
A short list that separates an audit-ready vendor from the rest: where our data is physically processed, whether the deployment can run on our infrastructure, whether data is isolated from other customers, what trail remains after each interaction, and what happens to the data when the relationship ends. If a vendor cannot answer plainly, that uncertainty is the answer.
In short
NIS2 does not require giving up AI. It requires choosing AI that gives you control over your data, runs inside your perimeter, and leaves a trail you can show an auditor. That is the difference between a tool that obstructs compliance and one that supports it. If you want to discuss how this looks deployed on your infrastructure, let's talk.
This is not legal advice. The scope of obligations depends on your organization's profile; if in doubt, consult a lawyer or compliance advisor.