Services · Building block

AI training and governance: trained, governed, documented.

AI training on real cases: your staff learn where AI helps and where it does not. Plus rules and evidence under the EU AI Act and GDPR.

Obligations at a glanceAs of October 2026
  1. DPA with the providerGDPR Art. 28done
  2. AI literacy supportedAI Act Art. 4in progress
  3. Transparency noticeAI Act Art. 50in progress
  4. Risk class assessedAI Act Art. 6open
  5. Works council involvedBetrVG s. 87open
AI rules1 page

Three tools approved.with DPA

No HR data in external tools.limit

New tool: ask IT.process

  • Trained on real cases

    Each department practises on its own work. What is learned there gets used day to day.

  • Rules on one page

    Which tools, which data, who decides when in doubt. Short enough to be read.

  • Obligations documented

    Who was trained on what and when is in the record. That is how you show what Article 4 of the AI Act asks for.

  • A reasoned choice of model

    Models chosen by data type, hosting location and cost per case, with no lock-in to one provider.

Where it gets stuck today.

  1. 29.1%

    AI is in use already, just not approved.

    Dispatch writes customer emails with a private account. Nobody allowed it, nobody banned it, and there is no data processing agreement.

    IW Cologne 2026: share of workplace AI users on tools their employer does not provide
  2. Unclear which obligations apply.

    AI literacy, transparency, high risk, GDPR: hardly anyone knows exactly which of it affects their own business. And nobody says what may go into which tool.

    A first assessment: the EU AI Act check
  3. 66%

    Skills are missing, and so is training.

    Most German companies rate their staff's AI skills as low, and a quarter do no training at all. One webinar for everyone changes little.

    Bitkom Research, AI in the German economy 2026, p. 13

Process

Training, rules, record.

Your team practises on real cases, the rules fit on one page, and the record shows who was trained on what and when. So AI at work is allowed, understood and documented.

Sales trainingCase 3 of 5
  1. Participant

    Summarise the utility's tender: deadlines and minimum requirements.

  2. Bid deadline 14 Nov, ISO 9001 required, penalty 0.5% per week of delay.

    Tender, pp. 4 and 9
  3. Trainer

    Good. Now we check the penalty clause against the original.

AI rulesDraft

Model choice

Customer data: model in a German data centre, no training on inputs.approved

Public texts: standard model with DPA.approved

HR data: no external tool.limit

When in doubt, the head of IT decides.

AI literacy recordAI Act Art. 4
  1. Sales training12 people, 14 Octrecorded
  2. Dispatch training9 people, 21 Octrecorded
  3. Management sessioncost, limits, dutiesrecorded
  4. Systems assessed3 systems, no high riskchecked

EU AI Act

Which deadlines apply to you.

Since the Digital Omnibus, in force since 27 July 2026, Article 4 asks for measures to promote the development of AI literacy. You do not have to guarantee a set level per person, but you should still be able to show the measures.

  • Since February 2025: literacy and prohibitions

    Applies to every business that uses AI at work, even if it is only a chat tool.

  • Since August 2026: transparency

    People talking to an AI are told so, under Article 50. For systems already on the market, labelling of generated content runs until 2 December 2026.

  • From December 2027: high-risk

    For example recruitment or creditworthiness under Annex III. High-risk in products such as machinery follows on 2 August 2028.

EU AI Act deadlinesAs of October 2026
  1. Art. 4 AI literacy, Art. 5 prohibitionssince 2 February 2025applies
  2. Art. 50 transparencysince 2 August 2026applies
  3. Labelling, systems already on the marketby 2 December 2026running
  4. High-risk under Annex IIIfrom 2 December 2027coming
  5. High-risk in productsfrom 2 August 2028coming
Article 4 in briefDigital Omnibus version

AI literacy

Businesses take measures to promote the AI literacy of their staff.required

No minimum level per person, no hours, no certificate.not required

Approaches

Which tool, which frame.

The tools differ less in quality than in where the data sits and what happens to it. Three questions per tool: is it trained on your input, where is it processed, and how do you show the measures to promote AI literacy under Article 4 of the EU AI Act.

  • Your own chat on European models

    Our default

    Our stack, run in Germany

    Fits when
    your team should work with AI and inputs should be processed in the EU. The chat runs on models in EU data centres, can draw on your knowledge system and states by itself that an AI is answering.
    Mind
    For pure Office work in Microsoft 365, Copilot can sit closer to daily work; we compare both on your cases.
  • Microsoft 365 Copilot

    Managed, in Microsoft 365

    Fits when
    your team works in Microsoft 365. Copilot sees mail, files and Teams with the rights of the person asking.
    Mind
    Microsoft states no training on your data and coverage by the EU Data Boundary; overly broad SharePoint shares suddenly become visible.
  • ChatGPT Business or Enterprise

    Managed, OpenAI

    Fits when
    a general assistant for writing, analysis and custom GPTs is needed, independent of Microsoft 365.
    Mind
    OpenAI states no training on business data; OpenAI offers data residency in Europe for Enterprise, not for Business.
  • Claude Team or Enterprise

    Managed, Anthropic

    Fits when
    long documents, contracts or code are at the centre. There are several comparable alternatives.
    Mind
    Anthropic states no training on data from its business plans; check data location and contract per plan.
  • Usage rule and AI register

    Governance, one page and one list

    Fits when
    Always. The rule says which tool is allowed with which data; the register lists tools, models and purposes for the EU AI Act.
    Mind
    Only as good as its upkeep: one person owns it, and new tools go in before they are used.
  • Training on real cases

    Enablement, with a record

    Fits when
    Always, and before a broad rollout. Practice runs on cases from your own work, not on examples from the internet.
    Mind
    Since the Digital Omnibus, Article 4 asks for measures to promote AI literacy; record who was trained on what and when.

Tasks and limits.

What inviqon takes on

  • Inventory: which tools, for what, with which data
  • Training on real cases, with a follow-up session
  • AI rules on one page, with a date and an owner
  • Model choice by data type, hosting location, cost per case and training use
  • EU AI Act assessment, notes for the GDPR record and impact assessment
  • A record of who was trained when on which tools and limits

What your team takes on

  • Management approves the tools
  • The works council co-decides where there is one (BetrVG s. 87)
  • Every AI statement used is checked against the original by the user
  • A named person maintains the rules and the record
  • Not included: legal advice and data protection impact assessment; we say where one is needed
  • Limit: no ban as a substitute, use just moves to private devices

AI training & governance

The AI strategy picks the processes. AI training & governance equips the people and sets the frame: who uses which tool with which data, and how you show it.

Which one when?

  • AI strategy

    Want to know first which processes are worth it? Then the AI strategy. Inventory and risk classification belong to both.

  • Pilot & implementation

    If we build a system for you, the training is part of the project anyway.

  • AI knowledge systems

    Anyone who gets answers with a source has to be able to check them. Knowledge system and training go together.

The way there.

From the first call to a system in operation. Each phase stands on its own; none ties you to the next.

  1. Intro call

    free

    30 minutes. We work out where AI fits for you and whether an AI strategy is worth it.

  2. AI strategy

    €1,990

    AI strategy for mid-sized companies: which of your processes suit AI, what the build costs and brings back, and which ones are not worth the effort.

  3. Pilot & implementation

    from €5,000

    AI pilot at a fixed price: one use case, fixed scope and acceptance criteria in writing up front. Built in your systems, tested with your data.

  4. Operation & improvement

    from 20% a year

    Running AI systems: hosting in Germany, monitoring, model and prompt maintenance, ongoing measurement, and the next use case on the same foundation.

Further reading

All posts

Frequent questions

Common questions about it.

The questions that almost always come up.

Is AI training for staff mandatory?

Yes, in the sense of Article 4 of the EU AI Act, but with no certificate requirement. Since February 2025, businesses that use AI at work must do something for their people's AI literacy. Since the Digital Omnibus took effect on 27 July 2026, that means taking measures to support the development of AI literacy. You do not have to guarantee a particular level per person, and there is no required number of hours or certificate. We record who was trained when on which tools and limits, so you can show it.

Does the EU AI Act apply to us if we only use ChatGPT?

Yes, as a deployer: Article 4 applies, and Article 50 depending on use. High-risk duties only cover certain purposes, such as screening job applicants. Our free EU AI Act check gives a first assessment.

Which EU AI Act deadlines apply?

Prohibitions and AI literacy since 2 February 2025. Transparency duties under Article 50 since 2 August 2026, with labelling of generated content for systems already on the market by 2 December 2026. High risk under Annex III, such as recruitment or creditworthiness, from 2 December 2027; high risk in products such as machinery from 2 August 2028. In Germany the Federal Network Agency supervises.

What belongs in an AI policy?

Approved tools, which data may go in, who decides when in doubt, that results are checked before use, and where to request a new tool. That fits on one page, and nobody reads more than one page.

Can we simply ban AI tools?

You can try. In practice use moves to private devices and accounts, and you lose sight of it entirely. Short rules with approved tools work better than a ban nobody can enforce.

Is the AI training a one-off or ongoing support?

Both, and the support is the part that works. Without the weeks after, three people carry on and the rest fall back. So we keep working on your real cases, not on exercises.

Which AI model should we use?

That depends on your data, the hosting location and the cost per case, not on this week's benchmark winner. We recommend per case and stay vendor independent.

The next step

Let us find out where AI pays off for you.

30 minutes, free and without obligation.