AI supervision in Germany: Federal Network Agency and KI-MIG
AI supervision in Germany has rested mainly with the Federal Network Agency (Bundesnetzagentur) since 29 July 2026: the Act on Market Surveillance and Innovation Promotion of Artificial Intelligence (KI-MIG) makes it the competent market surveillance authority for the EU AI Act, unless the act assigns another authority. It is also the single point of contact, the central complaints office and the competence centre for all other authorities. Exceptions apply to the financial sector (BaFin), AI in regulated products (the existing product authorities), media and public bodies of the federal states. For mid-sized companies this means one clear point of contact, free help with classification and an authority that follows up complaints. As of November 2026.
AI supervision under the KI-MIG: who is responsible for what
| Area | Competent AI supervisor | Legal basis |
|---|---|---|
| All AI systems without special competence, e.g. chatbots, AI in HR, transparency duties | Federal Network Agency | Section 2(1) KI-MIG |
| AI in products under Annex I Section A of the AI Act, e.g. medical devices or toys | the existing market surveillance for these products | Section 2(2) KI-MIG |
| AI of banks, insurers, payment and crypto service providers directly linked to a regulated financial activity | BaFin | Section 2(3) KI-MIG |
| High-risk AI in law enforcement, migration, border control, justice and democratic processes | independent AI market surveillance chamber at the Federal Network Agency | Sections 2(5) and 4 KI-MIG |
| Public bodies of the federal states | authorities under state law | Section 2(6) KI-MIG |
| Media for journalistic or advertising purposes | authorities under state law | Section 2(8) KI-MIG |
For most mid-sized companies the Federal Network Agency is therefore the AI supervisor. What the Regulation (EU) 2024/1689 itself requires is covered in our guide to EU AI Act obligations for companies.
What the AI supervision of the Federal Network Agency may do
Under Section 11 KI-MIG, market surveillance authorities hold the powers of the EU Market Surveillance Regulation 2019/1020 and the AI Act: requesting documents and information, examining systems and ordering measures. They may enter business premises during normal business hours, outside them only to prevent urgent threats to public safety and order, and may carry out certain checks remotely via programming interfaces.
- Single point of contact (Section 6): link to the European Commission's AI Office and other member states.
- Coordination and competence centre (Section 5): supports other authorities, promotes consistent interpretation and codes of conduct, involves business, unions and science.
- AI market surveillance chamber (Section 4): three members from the agency's leadership, independent, with an annual report to the Bundestag, first for 2026 by 31 March 2027.
AI supervision and complaints: who can complain
Any natural or legal person who suspects an infringement of the AI Act can complain (Art. 85 AI Act). Under Section 8 KI-MIG the Federal Network Agency runs the central complaints office and forwards complaints to the competent authority. According to its complaints page, it can also act on its own initiative; its decisions can be challenged before the administrative courts.
Fines under AI supervision: what infringements cost
This overview is not legal advice.
| Infringement | Maximum | Legal basis |
|---|---|---|
| Prohibited practices | 35 million euros or 7 percent of worldwide annual turnover | Art. 99(3) AI Act |
| Most other obligations, e.g. transparency | 15 million euros or 3 percent | Art. 99(4) AI Act |
| Incorrect or incomplete information to authorities | 7.5 million euros or 1 percent | Art. 99(5) AI Act |
| Missing explanation of a decision by high-risk AI and other national offences | 50,000 euros | Section 15 KI-MIG |
For SMEs, the lower of the fixed amount and the percentage applies to the AI Act amounts. No fines are imposed on authorities and other public bodies.
What AI supervision offers companies
- AI service desk: information on EU guidelines, codes of practice, AI literacy and duties per role.
- AI compliance compass: an online tool for a first assessment of an AI system's risk class.
- AI regulatory sandbox: under Section 13 KI-MIG the agency operates at least one sandbox where companies develop and test new AI systems with regulatory guidance; every member state must have one ready by 2 August 2027.
- Real-world testing: a test plan for a high-risk system outside a sandbox is deemed approved if the authority does not respond within 30 days (Section 14 KI-MIG).
AI supervision and associations: where to have a say
The competence centre must regularly involve business, unions and science and facilitate codes of conduct under Art. 95 of the AI Act. Under Section 19 KI-MIG the federal government evaluates the supervisory structure no later than 18 months after entry into force, i.e. by the end of January 2028, and again after three years, when companies' assessments of innovation-friendliness are expressly considered. How associations use AI themselves is shown on our page AI for associations.
Template: preparing for a request from AI supervision
| Document | Content | Reference |
|---|---|---|
| AI inventory | Which AI systems you use for what, provider, since when | basis for every question |
| Role per system | Provider or deployer, with a short reason | Art. 3 AI Act |
| Risk classification | Result of the compliance compass or own assessment | Art. 6 AI Act |
| AI literacy evidence | Training, participants, content, date | Art. 4 AI Act |
| Transparency | Chatbot notices, labelling of generated content | Art. 50 AI Act |
| Contact person | Who answers requests and decides | internal responsibility |
Labelling is explained in our guide to the AI labelling obligation. After the Digital Omnibus, the strict high-risk obligations apply only from 2 December 2027 (Annex III) and 2 August 2028 (Annex I).
Checklist: AI supervision in your company
- Competent supervisor identified
- AI inventory created and up to date
- Role and risk class recorded per system
- AI literacy training documented
- Art. 50 notices and labelling in place
- Contact person for authority requests named
- Agency offers checked: service desk, compass, sandbox
Questions about AI supervision
Who supervises AI in Germany?
Since 29 July 2026, the Federal Network Agency is generally competent under Section 2 KI-MIG. Exceptions apply to the financial sector (BaFin), AI in regulated products, media and public bodies of the federal states.
What is the KI-MIG?
The German Act on Market Surveillance and Innovation Promotion of Artificial Intelligence implements the EU AI Act in Germany. It sets the competent authorities, support for companies such as the AI regulatory sandbox and supplementary fine provisions.
Can I complain to the Federal Network Agency about an AI system?
Yes. Any natural or legal person can report a suspected infringement of the AI Act to the agency's central complaints office, which forwards it to the competent authority where needed.
Do companies have to register their AI systems with the Federal Network Agency?
For most systems, no. Registration is required only for high-risk systems and, in simplified form, for Annex III systems the provider classifies as not high-risk under Art. 6(3): usually in the EU database, and for critical infrastructure under Annex III point 2 in a non-public register kept by the agency.
Want to know how well your company or association is prepared for requests from AI supervision? Go through your AI inventory with us in a short introductory call. More from the agency: Federal Network Agency on AI.
Transparency notice: This article was created with the assistance of artificial intelligence.