The short answer
Usually not, for the model itself. But if you offer an AI system built on it under your own name, you are the provider of that system (Article 3(3)). For the general-purpose model, the Commission’s guidelines say that someone who modifies a model becomes its provider only when the change significantly alters the model’s generality, capabilities or systemic risk, and that few modifications today reach that point. Guidelines are not binding law, but they show how the Commission reads the rules.
The indicative criterion
The guidelines use compute as the indicator: the modification counts when it uses more than one third of the compute used to train the original model. If that figure is unknown, the guidelines use one third of 1025 FLOP for models with systemic risk, and one third of 1023 FLOP for other models.
If you do become a provider
Your duties cover the modification only. You add documentation about the changes, and your copyright policy and training data summary cover only the data used for the modification. If you modify a model with systemic risk beyond the threshold, the new model also has systemic risk, and you must notify the Commission.
High-risk systems are different
For high-risk AI systems, Article 25 says a distributor, importer, deployer or other third party becomes the provider if it puts its name or trademark on the system, makes a substantial modification so that the system stays high-risk, or changes the purpose of a system, including a general-purpose one, so that it becomes high-risk.
Document your fine-tuning data anyway
Whatever your role, keep a record of the data you fine-tuned on: sources, cleaning, personal data found and how it was handled. See how to prepare company data to fine-tune an LLM. This guide is general information, not legal advice.