EU AI Act consulting I help you put the EU AI Act into practice without turning it into a major project.
For SMEs and fintechs in Germany. We write down which AI you run, prepare the risk classification and compile the documentation. I built CompliKI, my own software, for exactly this work. This is not legal advice.
What the EU AI Act requires of your company
The EU AI Act, called KI-Verordnung in German, has been in force since August 2024 and applies in stages. It doesn't just affect companies that develop AI. If you only use AI, for example ChatGPT, Copilot or off-the-shelf software that pre-sorts job applications, you have obligations too.
How many depends on the risk of each system and on your role. The Act mainly distinguishes whether you offer an AI system under your own name (provider) or just use one (deployer). For most everyday office tools the effort is manageable. It gets demanding with high-risk AI, for instance when AI selects job applicants or assesses the creditworthiness of individuals.
In summer 2026 the EU postponed the obligations for high-risk AI. The prohibitions, the AI literacy requirement and the transparency obligations still apply. And you only know whether you use high-risk AI once you've written down which AI you run.
What applies, and from when (As of November 2026)
- since 2 Feb 2025Prohibitions and AI literacyCertain uses of AI are banned, for example emotion recognition in the workplace, except for medical or safety reasons. Companies have to do something so that their staff can handle AI (AI literacy), for instance through training or guidance. The Act doesn't require a certificate.
- since 2 Aug 2025Large AI models, supervision, finesRules for providers of general-purpose AI models such as those behind ChatGPT. Since then it has also been settled who supervises and which fines are possible.
- since 2 Aug 2026TransparencyWith a chatbot it must be clear that an AI is answering. That duty lies with the chatbot's provider (Art. 50(1)). If you only use a chatbot, check whether your provider meets it. Realistic AI-generated images, video and audio (deepfakes) must be labeled.
- from 2 Dec 2027High-risk AI in certain areas of useFor example AI in recruiting, credit scoring or education (Annex III of the Act). Originally August 2026, postponed in summer 2026. For systems already on the market or in service before then, the obligations are expected to apply only if their design is significantly changed afterwards (Art. 111(2)). The EU has not yet explained in detail what counts as a significant change.
- from 2 Aug 2028High-risk AI in regulated productsAI as a safety component, for example in medical devices or toys (Annex I). Originally August 2027.
This overview lists the stages that matter for most companies. It's a summary in my own words, not legal advice. What counts is the text of the regulation.
How I help with the EU AI Act
List your AI systems
We write down which AI you run: systems you built yourself, off-the-shelf tools such as Copilot or ChatGPT, and AI embedded in other software. For each system we note what it's used for, who is responsible for it and who supplies it.
What you getAn AI inventory your team can maintain on its own.
Prepare the classification
For each system we go through the same questions: what it's used for, who it affects and what your role is. That gives a provisional classification, from prohibited through high-risk to low risk. For each one I write down why. Whether it holds up legally and which obligations apply in detail is for your law firm or legal department to check.
What you getA traceable provisional classification per system and a list of questions for your lawyers.
Compile the documentation
Inventory and classification turn into the documentation you can present when customers, auditors or authorities ask. It states who is responsible, which rules apply to usage, how your staff are trained and what changed when.
What you getDocumentation you can present, and clear responsibilities.
How we work together
Step 1
Intro call
30 minutes, free. You tell me where you stand: whether a list of your AI systems exists, who has been looking after it so far and what's urgent. Then I'll tell you whether and how I can help.
Step 2
Inventory and classification
First you get a proposal with scope, timeline and price. If you agree, I go through your departments to find where AI is in use and prepare the classification of the systems with you.
Step 3
Documentation and handover
You get the inventory, the provisional classifications, the documentation and the list of open questions for your lawyers. I show your team how to keep it all up to date. Whether you use CompliKI for that or your own tools is up to you.
I built CompliKI, the software for exactly this
CompliKI is my software for the EU AI Act, made for small and mid-sized companies. It lets a company list its AI systems, classify the risk and export the documentation. To build it, I had to translate the regulation article by article into questions, rules and templates.
That showed me where it gets difficult: the question of what counts as an AI system in the first place, and whether a company is a provider or a deployer.
Before that I spent more than 15 years developing digital products, mostly in regulated industries: mobile financial services at Telefónica, later cross-border payments at PPRO Financial.
The EU AI Act for fintechs and banks
Financial companies such as banks, payment providers and fintechs already have plenty of rules to follow: supervisory requirements, outsourcing rules, DORA and the GDPR. The AI Act comes on top. An AI project passes through more hands there than elsewhere: the business, IT, compliance, data protection and sometimes internal audit.
The Act also classifies some typical use cases of the industry as high-risk, for example when AI assesses the creditworthiness of individuals. It explicitly exempts the detection of financial fraud. For companies supervised by BaFin, BaFin is responsible for AI directly related to the regulated financial business. For other AI in the same company, for instance in HR, it's the Federal Network Agency (Bundesnetzagentur).
I know from my own work how the business, IT, compliance and data protection need to work together. I plan for that from the start, so a project doesn't get stuck just before launch.
What I don't do: legal advice
I'm a computer scientist and product manager, not a lawyer. I help with the practical side: which AI systems you run, how they're likely to be classified, what you should document and how you organize this in day-to-day work.
The legal assessment of an individual case stays with your law firm or legal department. That applies above all when a system could be high-risk AI or when contracts with providers need review. I prepare the ground for your lawyers: the inventory, provisional classifications with reasons and a list of open questions. That way the legal review doesn't start from zero.
I also can't promise that an authority will agree with your classification. But in the end you know which AI you run and can explain how you handle it.
Frequently asked questions about the EU AI Act
Don't see your question? Send me a message.
EU AI Act, KI-Verordnung, KI Act: are they the same thing?
Yes. They all refer to Regulation (EU) 2024/1689. KI-Verordnung is the common German name, also used by the Federal Network Agency (Bundesnetzagentur). EU AI Act is the English name, KI Act a German short form. Lawyers in Germany often write KI-VO.
Does the AI Act apply if we only use ChatGPT or Copilot?
Yes. In the Act's terms that makes you a deployer. You support your staff's AI literacy with training or guidance, don't use AI for prohibited purposes, and label realistic AI-generated images, video or audio. It changes if you use the tool for a high-risk purpose, such as selecting job applicants. Then you may count as the provider of a high-risk system yourself, with considerably more obligations.
The AI Act was postponed. Do we still have time?
Only in part. The obligations for high-risk AI were postponed and now apply from December 2027 and August 2028. The prohibitions, the AI literacy requirement and the transparency obligations already apply. And to know whether you use high-risk AI at all, you first need a list of your AI systems.
Do we need an AI officer?
No. The Act doesn't require an AI officer. It still makes sense to have one person in charge: someone who keeps the list of AI systems, is the point of contact for the departments and notices when a new tool arrives. In a small company that can be someone from IT or data protection.
Is software like CompliKI enough, or do we need consulting?
It can be. CompliKI takes you step by step through inventory, classification and documentation, and it's built so that a company can do this on its own. Consulting is worth it when nobody has the time internally, when many departments are involved or when you work in a regulated industry. In the intro call I'll tell you which I'd recommend.
What does the consulting cost?
The intro call is free. After that the price depends on the scope: a company with five AI tools takes less time than one with fifty. Before you commit to anything, you get a proposal with scope, timeline and price.
Tell me where you stand with the EU AI Act.
30 minutes, free. Afterwards I'll tell you where I'd start.