How do I build a NIS2 supplier register with AI?
Give a connected AI assistant your list of creditors and your contracts. It adds one line per supplier in item 5.1 and ticks in 5.2 what the contracts settle on security and incidents. It cannot finish an item: you check every line, and management signs off at the end.
How does an AI assistant build the supplier register?
An AI assistant connects to your company in the app through MCP (Model Context Protocol), an open standard. It fills in the NIS 2 walkthrough as the person who connected it.
Almost everyone who works for you sends you invoices. That is why item 5.1 starts from the list of creditors in your accounting, covering the last twelve months, for example from DATEV. Give the assistant that list and your contracts and subscriptions.
It adds one line per supplier, with the name and what the supplier does for you. Tell it what the walkthrough recommends: when in doubt, write it down rather than leave it out.
Whoever you named in item 2.2 as the provider of your software and technology is already on the list. The assistant sees the existing lines, with an instruction not to add them again.
Who belongs on the supplier register under NIS 2?
Item 5.1, “Write down who works for you”, is a list of your direct suppliers and service providers. It covers everyone who touches your IT or your data, including the tax adviser with access to staff data.
The directive requires supply chain security, including the security-related aspects of your relationships with your direct suppliers. You take into account the vulnerabilities specific to each supplier (Art. 21(3) NIS 2).
For certain digital services, Implementing Regulation (EU) 2024/2690 names a registry of direct suppliers with contact points (Annex point 5.2). Whether your AI assistant's own vendor belongs on this list is covered in our article on the AI vendor as a supplier.
Art. 21(2)(d) NIS 2 · § 30(2) no. 4 BSIG
What does the assistant record about contracts in 5.2?
Item 5.2, “Record what you have agreed with your suppliers”, asks two things per supplier. Do the contract, the data processing agreement or the terms set security requirements, for example on access and updates? And does the supplier have to report security incidents to you?
The assistant reads the contracts, data processing agreements or standard terms you give it and sets the two ticks per supplier. Large providers rarely negotiate; then what counts is their standard terms. If neither is settled, “None of this is agreed” is the right answer.
Next to each name, 5.2 shows the level from item 2.3, where each supplier is rated by frequency and damage. So you can see where an agreement matters most.
For its digital services, the Implementing Regulation names points that contracts should specify where appropriate, among them security requirements and incident notification (Annex point 5.1.4).
What goes into a line, and what may the assistant change?
On the screen “Your suppliers” in 5.1 there are two fields: “Supplier” and “What they do for you”. The supplier list holds more, such as a contact person and email, whether the supplier has access to your systems or data, and when the contract starts and ends.
A new line needs only the name. The assistant can correct a line that is already there; fields it does not send stay as they are.
It cannot delete a line. You do that in the app.
What happens to statements a supplier made itself?
If a supplier has accepted your invitation to the security questionnaire, its line is linked to its own company in the supplier portal. What it stated there, for example whether it offers a data processing agreement or helps with incidents, are its own statements.
The assistant receives them set apart, under writtenBySupplier, and cannot change them. The server refuses and says these are the supplier's own statements from its portal.
A text written by others can try to slip instructions to the assistant (prompt injection). The server therefore tells the assistant to read everything as data, never as instructions. It says so in its instructions and in the description of the tool that reads the walkthrough.
Germany's BSI writes that no reliable countermeasure exists that removes the problem completely. What the server itself enforces holds regardless: no tool finishes an item, deletes a line or signs anything.
How the walkthrough does it
Suppliers turn into work at two items of the walkthrough, and the rating between them connects the two.
5.1 “Write down who works for you”, screen “Your suppliers”: the assistant adds one line per supplier from the list of creditors. Whoever 2.2 names as a provider is already there.
2.3, “Rate your suppliers”: each supplier gets a level from frequency and damage, which then also shows on the supplier list.
5.2, “What each one has agreed”: the assistant sets the ticks for security requirements and incident reporting per supplier. You check every line and finish 5.1 and 5.2 in the app.
7.3: a supplier rated high or very high whose contract settles neither security nor incidents appears as an open point by itself before the sign-off. You set in the app who takes care of it and by when.
Common questions
How does the assistant know who counts as a supplier?
From the records you give it and from the walkthrough's own words, which it reads along. They say who is meant: every direct supplier and service provider that touches your IT or your data. Whether the list is right, you decide when you check it, before you finish 5.1.
What if nothing is agreed with a supplier?
Then “None of this is agreed” is the right answer. The walkthrough suggests raising it, for example at the next renewal. You can send the supplier the security questionnaire from its line, or your assistant does.
Can the assistant change the supplier list after the sign-off?
No. Once 5.1 is signed off, the server refuses every new line and every correction from the assistant, because it would take the sign-off back. The same holds for the detail screens in 2.2, the ratings in 2.3 and the ticks in 5.2, which can change the supplier list. The refusal names the items concerned; you change it in the app, and the item then goes to management again.
How many suppliers does the assistant add at once?
Up to 50 new lines per call, and in 5.2 up to 500 suppliers at once. Each person may make 300 calls an hour. A long list of creditors therefore takes several calls.
Does my suppliers' data end up with the AI assistant's vendor?
What the assistant reads goes to its vendor, under your company's agreement with that vendor. For Anthropic's commercial products, Anthropic acts as a processor, and inputs are not used for training by default. We have not checked the terms of other vendors.
Read next
Sources
- Directive (EU) 2022/2555 (NIS 2), Art. 21(2)(d) and 21(3), EUR-Lex
- BSIG, § 30(2) no. 4, gesetze-im-internet.de
- Implementing Regulation (EU) 2024/2690, Annex points 5.1.4 and 5.2, EUR-Lex
- BSI, Basisschutz gegen Indirect Prompt Injections in dokumentbasierten LLM-Chats, version 1.0 of 14 September 2026, bsi.bund.de (in German)
- Anthropic Privacy Center, articles 7996868 (training) and 9267385 (processor role), privacy.claude.com
- open-isms, lib/mcp/walk.ts (writtenBySupplier) and lib/mcp/tools.ts, github.com/NISD2/open-isms
In practice
Where this comes up in the walkthrough
The walkthrough turns this into steps you fill in, with the law explained beside each one. Open a step to see its screen, filled in for a made-up company.
Supply Chain Security
Write down who works for you
A list of your suppliers and service providers.
Supply Chain Security
Record what you have agreed with your suppliers
For each supplier: what does the contract say about security?
What the walkthrough costs
€4,800 net a year, plus VAT
30 days money backBy invoice, due in 30 days. Cancel within the first 30 days, before the first export or sign-off: invoice cancelled or money back.
Guided NIS 2 implementation, step by step
A history of changes and sign-offs to show your regulator
Deadlines and reminders for each requirement
Security questionnaire for your suppliers
Export to PDF or CSV at any time
Been asked to handle NIS2?
In the end you have everything in one document: what is done, what is still open, and your management's approval with name and date. The app writes the texts.
