
Self-Learning AI: Which Systems AI Funding MV Recognises
Image: Fatemeh Rezvani / Unsplash
The AI Funding MV only funds projects that predominantly use self-learning AI – processes based on machine learning and neural networks. According to TBI, this explicitly includes language models. Purely rule-based or deterministic systems are not eligible for funding; proof is provided by the manufacturer or retailer.
The AI Funding MV only supports projects that predominantly self-learning AI. According to No. 4.1 of the Directive of 10 June 2026 Processes based on machine learning and neural networks that independently recognize patterns, correlations or decision-making rules from data. The TBI clarifies: This also includes systems that have learned in development and no longer learn in use –"e.g. language models".Purely rule-based or deterministic systems without the ability to learn are not eligible.
This demarcation is the most important hurdle in terms of content of the program. Many products advertise with "AI", but work with fixed rules. If you clarify this before the application, you save yourself a rejection.
What the Directive means by self-learning AI
The wording of the guideline: The object of funding must "predominantly include the use of artificial intelligence in the sense of self-learning systems. This includes methods based on machine learning and neural networks, through which the system is able to independently recognize patterns, relationships or decision-making rules from data and independently optimize its performance over time."
The information for TBI applicants complements the decisive sentence for practice: Systems that have learned themselves in the development phase and no longer learn in funded use also count – for example, language models. This covers most of today's generative AI applications.
Eligible or not? Typical cases
System | Self-learning in the sense of support? | Justification |
|---|---|---|
AI telephone assistant with voice model | Yes | Speech recognition and conversation management are based on neural networks |
Website chatbot with language model | Yes | Language model; website chatbots are explicitly mentioned |
Document Search with Language Model (RAG) | Yes | Knowledge and support system based on a language model |
Image recognition for quality inspection | Yes | trained neural network |
Forecasting model for sales or maintenance | Yes | machine learning on historical data |
Workflow automation with fixed if-then rules | no | deterministic, without the ability to learn |
Click chatbot with fixed answer tree | no | rule-based |
Telephone menu with button selection | no | rule-based |
The table classifies typical system types; the specific product and the evaluation of the TBI are decisive. An automated process chain can still be eligible for funding if the majority of it consists of AI – for example, if a language model understands and assigns incoming emails and fixed rules only do the rest.
The AI certificate in the application
The application includes proof of the use of artificial intelligence. According to TBI, the manufacturer or distributor of the software must provide it. Ask your provider for a written confirmation that describes:
- which AI methods are used (e.g. language model, neural network for image recognition),
- what proportion of the solution is based on these methods,
- what the AI is used for in the specific project.
In the case of solutions consisting of several building blocks, it should be clear that the AI share predominates. Pure integration or automation work without an AI core is not enough.
Two further AI obligations under the Directive
- Identify AI texts in the application: Content of the project description generated with AI must be identified (No. 6.7).
- Comply with the EU AI Regulation: The funded AI must comply with the provisions of the EU AI Regulation (No. 6.8) – such as the obligation of transparency when people speak or write to an AI system.
What this means for our solutions
Borderline case: Standard software with AI functions
Many well-known programs have now built in AI functions – office packages, accounting or industry software. This is where it gets difficult: According to TBI, standard software is not eligible for funding if it is procured on its own. An office subscription with AI assistant is therefore not a typical funding project. The situation is different when an AI solution is introduced, adapted and connected for a specific process in the company. Clarify borderline cases with TBI before applying.
Questions to ask your provider
- What AI methods is the solution based on – language model, image recognition, forecasting model?
- What part of the solution is AI, which is integration or fixed rules?
- Do you issue written proof of AI for the funding application?
- Where do the models run, and where is the data processed?
- How is the transparency obligation of the EU AI Regulation implemented?
- Can the license be offered with a three-year term and upfront payment?
Our AI Phone Assistants,AI chatbots, which AI Document Search and AI Agents are based on language models; we issue the AI proof for the application. A pure Process automation with fixed rules, on the other hand, is not eligible for funding on its own – but can be part of a project whose core is an AI solution. What a Language model explains the glossary; the key data of the funding can be found on the funding page AI Funding MV.
Frequently Asked Questions
What is self-learning AI in the sense of AI Funding MV?
Methods based on machine learning and neural networks that independently recognize patterns or decision rules from data. According to TBI, this also includes language models that have learned during development.
Are applications with ChatGPT-like language models promoted?
Yes, language models are considered self-learning systems. However, only the introduction of an AI solution in the company that covers one of the areas of application of the directive is funded.
Is workflow automation eligible?
Purely rule-based automation is not. A project can be eligible for funding if its majority is based on AI, such as a language model that understands documents.
Who confirms that my system is AI?
The manufacturer or distributor of the software provides proof of the AI characteristics; it is submitted with the application.
As of November 2026, the guideline on the granting of de minimis grants to promote the use of artificial intelligence of 10 June 2026 (Official Gazette No. M-V 2026 p. 521) as well as the information provided by the TBI are decisive. This article is not legal or tax advice.

Muhamed Alahmed
With over 10 years’ experience in IT, I develop solutions that not only work from a technical perspective, but also create real added value and open up new possibilities.
More about bettersorted →More articles

AI Funding MV: Which Costs Are Eligible – and Which Are Not
What AI Funding MV pays for: hardware, software, consulting, training – and what it doesn’t: standard IT, short licences, own labour. With example.

Getting an AI Phone Assistant Funded: 50% Grant in Mecklenburg-Western Pomerania
AI phone assistant with a 50% grant: how businesses in Mecklenburg-Western Pomerania use AI Funding MV – requirements, costs, application.

Getting an AI Chatbot Funded: 50% Grant in Mecklenburg-Western Pomerania
Website chatbot with a 50% grant: how companies in Mecklenburg-Western Pomerania get an AI chatbot funded through AI Funding MV.
Stay up to date on AI topics
Short updates on AI automation, funding programs and new posts — no spam, unsubscribe anytime.