AI Act: provider or deployer? When your organisation becomes a provider without noticing
Short answer
Whoever uses an AI system is a deployer; whoever develops it or places it on the market under its own name is a provider, with far more obligations. Article 25 of the AI Act turns a deployer into a provider if it offers a high-risk system under its own brand, substantially modifies it, or repurposes another system so that it becomes high-risk. For high-risk uses under Annex III these obligations apply from 2 December 2027.
The two roles
| Provider (Art. 16) | Deployer (Art. 26) | |
|---|---|---|
| Who | Develops an AI system or has it developed and places it on the market under its own name | Uses an AI system under its own authority, in a professional context |
| For high-risk | Risk management, data governance, technical documentation, conformity assessment, CE marking, registration, post-market monitoring | Use according to instructions, human oversight, keep logs, inform workers, report incidents, sometimes a fundamental rights impact assessment |
| Always | AI literacy of staff (Art. 4) and transparency towards people who interact with AI (Art. 50) | |
The baseline obligations for every organisation are in The AI Act in 2026: what your organisation must have in place; this article is about the change of role.
The three triggers in Article 25
- Own name or trademark. You offer a high-risk system that is already on the market under your own name, for instance as a module of your own software. You can divide the obligations with the original provider by contract, but externally you are the provider.
- Substantial modification. You change a high-risk system in a way that was not foreseen in the original conformity assessment and that affects the requirements or the intended purpose. It remains high-risk, but now under your responsibility.
- New intended purpose. You take a system that was not high-risk, including a general-purpose model, and use it for a purpose in Annex III: recruitment, creditworthiness, access to services, education. Nothing changes in the system, everything changes in your role.
The third case is the most common and the easiest to miss. A general chat assistant that HR uses to shortlist applications is a high-risk system, and whoever set up that use is its provider.
What applies if you run your own model?
Running an open-weight model on your own node does not make you a provider of a general-purpose AI model: the provider is whoever develops the model and places it on the market, such as Meta or Mistral. Only someone who substantially modifies such a model, for instance by fine-tuning it on their own data, can become a provider for the modified part under the Commission’s guidelines. Using a model unchanged and connecting it to your own documents is deployment, not provision.
If you build your own application from the model and use it for a high-risk purpose, Article 25 applies again: then you are the provider of that system. See Comparing open-weight models for the model side.
How to check your role
- Inventory. Which AI systems are running, who set them up, what are they actually used for? An AI register is the foundation.
- Compare against Annex III. Does a use touch employment, education, credit, essential services, law enforcement, migration or critical infrastructure?
- Read the vendor’s intended purpose. Is your use in the documentation? If not, you may be the first to define that purpose.
- Record changes. Assess every adjustment to a supplied system for substance beforehand and document it.
- Contracts. Agree with the vendor who carries which obligations on repurposing, and whether the vendor excludes conversion into a high-risk system.
The timeline after the Digital Omnibus
Regulation (EU) 2026/1744 moved the deadlines for high-risk systems: Annex III uses apply from 2 December 2027, AI in products under EU product safety law (Annex I) from 2 August 2028. Not moved are AI literacy (since February 2025), the obligations for general-purpose models (since August 2025) and the transparency duties of Article 50 (since August 2026). Anyone setting up a system for an Annex III purpose today therefore has time, but the role is already fixed. For training see AI literacy.
How werqly fits in
With werqly you run unchanged open-weight models on a node of your own, connected to your documents; that is deployment, not provision. werqly AI Control records which role uses which sources and for what, so that repurposing for a high-risk use does not happen unnoticed, and provides the log that deployers must keep.
Frequently asked questions
Questions on this topic
Are we a provider if we use ChatGPT or Copilot?
Normally not: you are a deployer. You become a provider if you set the tool up for a high-risk purpose under Annex III, such as shortlisting applications, or offer it under your own name as a high-risk system.
Does fine-tuning an open-weight model make us a provider?
It can make you the provider of the modified model, limited to the modification under the Commission’s guidelines. Running a model unchanged and connecting it to your own documents, by contrast, is deployment.
What is a substantial modification?
A change after placing on the market that was not foreseen in the original conformity assessment and that affects compliance with the requirements or the intended purpose. Whether that applies depends on the case; document every adjustment.
From when do the provider obligations for high-risk systems apply?
For Annex III uses from 2 December 2027, for AI in regulated products from 2 August 2028, after the postponement by Regulation (EU) 2026/1744. AI literacy and transparency duties already apply.
Sources
- AI Act — Regulation (EU) 2024/1689, Article 25
- European Commission — AI Act Service Desk, Article 25
- European Commission — Regulatory framework for AI (Digital Omnibus)
This article is general information, not legal advice. Rules and guidance change; check the sources listed or ask your lawyer or DPO if in doubt.