The picture we build of your company
Not a questionnaire sent to department heads. We look at what is actually running.
AI inventory
| System | Finding | Status |
|---|---|---|
| Microsoft 365 Copilot | 120 licences, approved by IT | approved |
| ChatGPT via private accounts | 37 employees, quotes and customer data pasted in | not approved |
| Chatbot on the website | not labelled as AI | required since 2 Aug 2026 |
| CV screening in the HR system | high-risk under Annex III | deadline 2 Dec 2027 |
| 6 AI features inside existing software | switched on quietly by vendor updates | not assessed |
Illustrative of a typical finding. Your result comes as an inventory, a risk assessment and an action plan.
Why this matters now
In the summer of 2026 the EU pushed the strict obligations for high-risk AI back to December 2027. What stuck in most boardrooms was a single sentence: there is time.
The risk was never the deadline. The risk is the operation.
- Data leaves the building — quotes, costings, personnel files, pasted into a chat window whose provider nobody has vetted.
- Cost with no cost centre — AI subscriptions on company cards, billed per token, nobody consolidating.
- Output with no review — AI text goes to customers without anyone signing off on it.
- Responsibility stays at the top. Not with your IT provider.
What applies today — and what was actually postponed
Most summaries online have been wrong since the Digital Omnibus. This is the position we work from.
- 17 Oct 2024national law NIS2 Directive — Cybersecurity duties, applied through each country's own law
- 2 Feb 2025in force Art. 5 AI Act — Prohibited AI practices
- 2 Feb 2025in force Art. 4 AI Act — AI literacy of staff
- 2 Aug 2026in force Art. 50 AI Act — Transparency for chatbots, AI content and deepfakes
- 2 Dec 2026coming Art. 50(2) — Machine-readable marking for existing systems
- 2 Dec 2027postponed Annex III — Standalone high-risk systems
- 2 Aug 2028postponed Annex I — Embedded high-risk AI
As at 25 September 2026. This is not legal advice — it tells you what to raise with your lawyer.
NIS2 across the EU
NIS2 is a directive, so it binds companies only through national law. The duties are broadly the same everywhere. The authority, the registration portal, the reporting route and the fines are not.
- In force 23
- Applies from 1 Oct 2026 1
- No national law yet 3
- Austriaapplies from 1 Oct 2026
- Belgiumin force
- Bulgariain force
- Croatiain force
- Cyprusin force
- Czechiain force
- Denmarkin force
- Estoniain force
- Finlandin force
- Franceno national law yet
- Germanyin force
- Greecein force
- Hungaryin force
- Irelandno national law yet
- Italyin force
- Latviain force
- Lithuaniain force
- Luxembourgin force
- Maltain force
- Netherlandsin force
- Polandin force
- Portugalin force
- Romaniain force
- Slovakiain force
- Sloveniain force
- Spainno national law yet
- Swedenin force
A group with sites in several countries answers to several laws at once. A German parent with plants in Poland and Spain has one law in force since December 2025, one since April 2026, and one that does not exist yet. The inventory maps this per site.
France, Ireland and Spain were referred to the EU Court of Justice in July 2026 for not transposing NIS2. We re-check the status before every engagement.
Three steps. Each one bookable on its own.
Each one ends in something you can hold in your hand.
AI inventory
We find out which AI is actually running in your company — technically, not by questionnaire.
- Inventory of every AI system and AI feature
- Data flows and provider locations
- Classification under the AI Act and the NIS2 law of every country you operate in
- Action plan ranked by effort and effect
- A paper your board or advisory board can work from
Establish control
The findings become a state of affairs you can stand behind — without taking AI away from your people.
- An approval process and AI policy that survives daily use
- One controlled route in, instead of private accounts
- Logging, access rights, labelling
- Staff training with evidence (Art. 4)
- Technical implementation with vendors we have vetted
Managed AI Control
The ongoing service. The name of the tier, not of the category.
AI use changes monthly. Control you establish once is gone again within a quarter.
- Ongoing watch for new AI systems in the business
- Consumption and cost by department
- Incident handling and reporting deadlines
- A quarterly report in your management's language
- An evidence file that survives an audit
Who we work with
A fit
- Companies between 100 and 10,000 employees across Europe
- Groups with sites in several EU countries — the same AI obligations everywhere, a different NIS2 law in each
- Manufacturing, engineering, logistics, healthcare, energy, retail — wherever NIS2 already bites
- IT leaders who know AI is in the building but have no handle on it
- Managing directors who want to use AI and secure it, not slow it down
- Companies that must keep data in the EU — by contract or by conviction
Not a fit
- Corporations with a staffed AI governance function. They need suppliers, not a partner.
- Companies who want a PDF for the auditor and no change otherwise. You can buy that cheaper.
- Businesses under 100 employees with nobody responsible for IT. You need the basics first.
- Anyone shopping for a platform. We are not a product with a login; we are accountability with a face.
100 – 500 employees
The managing director decides, usually with no dedicated security lead. Entry through the fixed-price inventory.
500 – 2,000 employees
IT leadership or the CISO drives it, management approves. Full assessment across sites, then control kept running.
2,000 – 10,000 employees
Security, legal and procurement are all at the table. Evidence for group audit and customer audits, subsidiaries included.
Independent — and we tell you where we earn
TOKAVE is owned by no vendor. We work with selected providers because control over AI does not work without technology. On some of them we receive a referral commission.
So we do it this way: before any recommendation, you learn whether we earn from it. The inventory is vendor-neutral and does not get cheaper or more expensive depending on what you buy afterwards. If the best answer for you is the one we earn nothing on, that is the one we recommend.
That is not a posture, it is the business model. We want to be the number the Mittelstand calls about controlled AI use — and you do not hold that position for two years by selling commissions.
The people who sign your invoice
You will not be handed juniors applying a method they learned last week.
25 years in B2B sales and partner business, including Adobe and Apple. Built ecosystems with Accenture, HCL and Infosys.
In IT security on the vendor side since 2008 — Palo Alto Networks, VMware, McAfee, Cisco, Datto. Advises TOKAVE on strategy.
What we hear most often
The AI Act was postponed, wasn't it?
Only in part. What moved were the obligations for high-risk systems. The prohibitions, the duty to ensure AI literacy among your staff and, since 2 August 2026, the transparency obligations all apply and are enforceable. And NIS2 is already national law in 23 of 27 EU member states, including Germany since December 2025. Austria follows on 1 October 2026.
Our IT provider handles that.
Your IT provider runs your systems. They do not decide which AI your sales team may use, they do not classify a high-risk system and they do not carry your organisational duties. Ask them for your AI inventory. If they deliver it, you do not need us.
We already have ISO 27001.
That is a good starting position — your management system exists. It does not cover which AI systems are in use, how they are classified, how output is reviewed or how you evidence competence. We build on your ISMS rather than a second one beside it.
What does it really cost?
The inventory has a fixed price you know before we start. After that you decide. We do not work on open day rates and we do not write a proposal before we know what is running in your company.
Does our data stay in Europe?
Yes. We work with processing inside the EU, and where a provider cannot commit to that cleanly, it does not reach our recommendation. Proving where your data actually flows today is part of the inventory.
How quickly do we see something?
First call within a week. First results and immediate measures after one week, the full inventory in four weeks. You do not wait for the final report.
Thirty minutes. No sales pitch.
We go through the AI systems you suspect are there, the obligations that apply to you today and what that means for your next six months. You get a written assessment afterwards — even if we never work together.