Article 25: when using AI makes you its provider

Under Article 16 of the AI Act, the provider of a high-risk AI system has to make sure it meets the requirements in Articles 9 to 15. Article 3 defines a provider as the person or body that develops an AI system, or has one developed, and places it on the market or puts it into service under its own name or trademark. A deployer is the person or body using an AI system under its own authority, other than in a personal non-professional activity.

Buy a tool, use it at work, and you are a deployer. You stay a deployer unless one of the three routes in Article 25 applies to a system you use.

Article 25 is the provision that moves the line. It converts a distributor, importer, deployer or other third party into a provider in three defined circumstances. Once converted, you are subject to the obligations in Article 16, which are the developer's obligations, not the buyer's.

None of this bites before 2 December 2027. That is when the high-risk rules for Annex III uses start to apply, under Regulation (EU) 2026/1744. The question is still worth answering now, because the answer changes what you buy and what you sign before then.

Article 25(1) states that a distributor, importer, deployer or other third party is considered a provider of a high-risk AI system, and subject to the provider obligations under Article 16, in any of three circumstances.

The first is branding: you put your name or trademark on a high-risk AI system that has already been placed on the market or put into service. Article 25(1)(a) does not override a contract that shares out the obligations differently.

The second is a substantial modification. You change a high-risk system already on the market, and it stays high-risk under Article 6. Article 3 counts a change as a substantial modification only if the provider did not foresee or plan it in its initial conformity assessment (its original check that the system meets the rules). The change must also either affect whether the system meets the requirements, or change its intended purpose.

The third is the one to read twice. You take an AI system that is already on the market and was not classified as high-risk, and change its intended purpose so that it becomes high-risk under Article 6. This includes a general-purpose AI system.

Note what the third route does not require. No code. No involvement from the supplier. No contract. No sale: putting a system into service for your own use counts. It turns on what the system is now for.

Say a business gives a general AI assistant a prompt to score each CV for a role out of ten against a list of criteria, with one line of reasoning. The scores go into a spreadsheet, and nobody reads the CVs in the lower-scoring half.

Annex III point 4 covers AI systems intended to be used for the recruitment or selection of natural persons, in particular to place targeted job advertisements, to analyse and filter job applications, and to evaluate candidates. That is a plain description of what the spreadsheet does.

The model's provider placed a general assistant on the market. Its stated intended purpose was not candidate evaluation. Using it that way changes the purpose and brings the system inside Annex III. On the face of Article 25(1)(c), the business using it that way is now the provider.

No software was developed and nothing new was bought. Once the high-risk rules apply, the change of use alone makes the business the provider of a high-risk AI system under Article 16. Because it also uses the system, it carries the deployer duties in Article 26 as well.

Article 26 is a manageable list. A deployer of a high-risk system has to use it as the instructions for use say. It has to give human oversight to a person who is competent and trained, and has the authority and support to do it. Where it controls the input data, it has to make sure that data is relevant and representative enough. It has to monitor how the system runs, report serious incidents, and keep the logs the system generates for at least six months. Before putting the system into use at work, it has to tell workers' representatives and the workers affected. And it has to tell people when they are subject to a decision made with it.

Article 16 is a much bigger job. It requires the system to meet every requirement in Chapter III Section 2. That means a risk management system running across the system's whole life (Article 9), and rules for the data used to train, validate and test it (Article 10). It means technical documentation written before the system goes to market and kept up to date (Article 11), automatic logging (Article 12), and instructions for use for the deployers who use it (Article 13). It also means human oversight designed into the system (Article 14), and accuracy, robustness and cybersecurity (Article 15). On top of that come a quality management system under Article 17, conformity assessment before the system is placed on the market or put into service, an EU declaration of conformity, CE marking, and registration in the EU database under Article 49.

One of those points back at you. Article 13 requires a provider to give deployers instructions for use. If you became the provider of a system you use in your own business, you are writing those instructions for yourself.

Failing to meet the high-risk obligations can bring administrative fines of up to 15 million euro or 3% of total worldwide annual turnover, whichever is higher, under Article 99. For a small or medium enterprise, Article 99(6) makes the cap whichever of the two is lower. For any business turning over less than 500 million euro, 3% of turnover is the lower of the two. Article 99(6a), added by Regulation (EU) 2026/1744, applies the same lower-of-the-two rule to small mid-cap enterprises.

Article 25(2) decides whether the original provider has to help you meet Article 16.

Where one of the three conversions happens, the provider that first placed the system on the market stops being the provider of that specific system. It must then cooperate closely with the new provider. It must make available the information, the technical access that can reasonably be expected, and the other help needed to meet the obligations in the Regulation, in particular for the conformity assessment of high-risk systems.

Regulation (EU) 2026/1744 amended Article 25(2) to spell out that this help includes, where relevant, technical documentation sufficient to assess compliance with the requirements in Article 16, information about known limitations and failure modes, and targeted technical access for testing and validation. A provider that owes this help and does not give it can be fined under Article 99(4).

That is a useful right. The technical documentation Article 16 requires includes a description of the training data, and of how the system was validated and tested and with what results. That is a record of work the original developer did.

The last subparagraph takes it away. Article 25(2) does not apply where the initial provider has clearly specified that its AI system is not to be changed into a high-risk AI system. In that case the provider does not have to cooperate with the new provider or hand over the documentation.

The exclusion only needs the supplier to have clearly specified that its system is not to be changed into a high-risk AI system. The Act does not say where the supplier has to specify this, which means a clause in the terms of use or usage policy may be enough. Since Regulation (EU) 2026/1744, a supplier that owes Article 25(2) help and does not give it can be fined under Article 99(4). If the exclusion is there and you convert anyway, you owe technical documentation about a system you cannot see inside.

Article 25 can only make you the provider of a system that is high-risk under Article 6, either through a use listed in Annex III or as a safety component of a product covered by Annex I.

Using a tool the way its supplier intended, however heavily, is use and not modification. Configuring options the provider documented, adding your own selection criteria inside a recruitment product, or running ten times the volume does not trigger Article 25(1)(b). The test in Article 3 has two parts. A change is a substantial modification only if the provider did not foresee or plan it in its initial conformity assessment. It must also affect whether the system meets the requirements, or change the intended purpose it was assessed for.

Article 25(1)(c) only converts you if the system becomes high-risk under Article 6. A use that is not listed in Annex III, and does not make the system a safety component of a product covered by Annex I, cannot convert you. Drafting marketing copy, summarising meetings, writing first-draft code, answering support email and translating documents are not Annex III uses. No amount of prompting turns them into one.

Article 6(3) goes further. Even a system whose use is listed in Annex III is not high-risk if it does not pose a significant risk of harm to health, safety or fundamental rights. That includes a system that does not materially influence the outcome of a decision. The exemption covers systems that do a narrow procedural task, improve the result of work a person has already completed, detect patterns in decisions without replacing or influencing the human assessment, or do preparatory work.

There is an exception to that exception, and it is why the CV example is hard to escape. Under Article 6(3), a system listed in Annex III that profiles people is always high-risk. Scoring a person against criteria to judge their suitability is profiling as Article 4(4) of the GDPR defines it, and the AI Act uses that definition.

The answer is decided system by system, not organisation by organisation. The same company can be the deployer of a payroll tool and the provider of a CV sifter it assembled itself.

Four questions settle it, and none of them needs a solicitor to start.

The high-risk obligations apply from 2 December 2027 for Annex III uses such as recruitment, and from 2 August 2028 for AI in products covered by Annex I, following the Digital Omnibus on AI, Regulation (EU) 2026/1744. No AI Act fine can attach to the spreadsheet before 2 December 2027. The GDPR applies already. Article 22 gives applicants the right not to be subject to a decision based solely on automated processing that significantly affects them. Recital 71 gives e-recruiting without any human intervention as an example.

The reason to look now is your supplier contracts. A contract still running on 2 December 2027 will be in force when the high-risk rules apply. The clause that decides whether the supplier has to help you under Article 25(2) belongs in the contract you sign before then.

A workflow is easier to change before it is embedded. Article 111(2) keeps a high-risk system that was placed on the market or put into service before 2 December 2027 outside the Regulation until its design changes significantly. The Act does not say how that rule treats a workflow converted under Article 25(1)(c) before that date, so do not count on it.