If your company uses third-party AI systems and adapts them, puts its brand on them or changes what they are for, this concerns you. Five articles in this series say the same thing in different words: the Article 17 quality management system, the Article 18 documentation retention duty, Annex IV, the Article 72 post-market monitoring duty and the Article 73 serious incident reporting duty are not obligations of the deployer — with one caveat for Article 73: if the deployer of a high-risk system identifies a serious incident and is not able to reach the provider, Article 26(5) applies Article 73 to it mutatis mutandis.
This one closes the series by saying the opposite: when they are. Because there are three circumstances in which the deployer is considered the provider of a high-risk AI system, and with that change, all the previous ones land on them.
The three scenarios, closed list
Article 25(1) is explicit: any distributor, importer, deployer or other third-party shall be considered to be a provider of a high-risk AI system — and shall be subject to the obligations of Article 16 — in any of these circumstances, all three confined to high risk:
“(a) they put their name or trademark on a high-risk AI system already placed on the market or put into service, without prejudice to contractual arrangements stipulating that the obligations are otherwise allocated”
“(b) they make a substantial modification to a high-risk AI system that has already been placed on the market or has already been put into service in such a way that it remains a high-risk AI system pursuant to Article 6”
“(c) they modify the intended purpose of an AI system, including a general-purpose AI system, which has not been classified as high-risk and has already been placed on the market or put into service in such a way that the AI system concerned becomes a high-risk AI system in accordance with Article 6”
In paragraph 1 there are three, and there is no fourth. (Paragraph 3 adds the case of the manufacturer of a product covered by Section A of Annex I whose high-risk AI system is a safety component of the product and is placed on the market together with it, or put into service afterwards, under the manufacturer's name or trademark.) Integrating, customising, configuring or using something heavily isn't enough. The grey area many people imagine isn't in the article.
The figure walks through the three circumstances in Article 25(1) in order, after a prior question: is the system high-risk, or does the change make it so? If it is and any of the three applies — putting your name or trademark on a high-risk system already placed on the market or put into service, substantially modifying it so that it remains high-risk, or changing the intended purpose of one that wasn't high-risk so that it becomes so — you are considered the provider of the high-risk system, subject to the provider obligations in Article 16. If not, none of them applies, and provider status is assessed under the definition in Article 3, point (3).
Letter (b) doesn't mean 'any change'
It's worth pausing here, because "substantial modification" sounds like an everyday phrase, and it has its own definition in Article 3, point (23):
“a change to an AI system after its placing on the market or putting into service which is not foreseen or planned in the initial conformity assessment carried out by the provider and as a result of which the compliance of the AI system with the requirements set out in Chapter III, Section 2 is affected or results in a modification to the intended purpose for which the AI system has been assessed”
Two conditions narrow the scenario considerably. The first: the change must not have been foreseen in the initial conformity assessment. Configuring parameters the product offers, activating modules the manufacturer already contemplated, or adjusting thresholds within what was foreseen doesn't count if it was foreseen in that assessment, because the provider already assessed it.
The second: the change must affect compliance with the requirements in Chapter III, Section 2, or change the intended purpose for which the system was assessed. A change that touches neither of the two isn't substantial, however much work it took.
Put the other way round: substantially modifying a system means doing something to it that its manufacturer didn't contemplate and that alters what it was assessed for. It isn't customising it. And for letter (b) that isn't enough: the system must be high-risk and remain so after the change.
Letter (c) is the easiest one to cross without noticing
And it doesn't require touching the software.
It's enough to modify the intended purpose — the use for which the provider intended the system, as specified in its instructions for use, promotional or sales materials and statements, and technical documentation — of a system that wasn't high-risk and was already on the market or in service, so that it becomes high-risk in accordance with Article 6. The tool is the same; the intended purpose is different; the classification changes — and whoever changed the purpose becomes the provider.
It's exactly the question that stayed open when we talked about staffing agencies: a firm that offers its clients, as its own service, a system built on third-party tools may fall under letter (a) or letter (c) — if the system is, or becomes, high-risk — without ever having explicitly decided which. And the answer isn't in the use: it's in under what brand the service is provided and what it's said to do.
What happens to the initial provider, which is what closes the split
Article 25(2), as amended by Regulation (EU) 2026/1744, answers the question the series left hanging, and it does so bluntly:
“the provider that initially placed the AI system on the market or put it into service shall no longer be considered to be a provider of that specific AI system for the purposes of this Regulation”
“That initial provider shall closely cooperate with new providers and shall make available the necessary information and provide the reasonably expected technical access and other assistance that are required for the fulfilment of the obligations set out in this Regulation, in particular with regard to compliance with the conformity assessment of high-risk AI systems”
So the role transfers, it doesn't duplicate. Whoever falls into any of the three circumstances keeps the obligations; the original provider exits, and in exchange is bound to cooperate and, where relevant, to make available technical documentation sufficient to assess compliance with the requirements laid down in Article 16, inform the new providers about known limitations and failure modes, and provide targeted technical access, including for testing and validation.
With one caveat worth knowing before signing anything:
“This paragraph shall not apply in cases where the initial provider has clearly specified that its AI system is not to be changed into a high-risk AI system and therefore does not fall under the obligation to cooperate with the new providers and hand over the documentation.”
In other words: if the initial provider clearly specified that its system is not to be changed into a high-risk one, whoever changes it takes on the role without the initial provider being bound to cooperate with them or hand over the documentation. They're left with the provider's obligations and without that support to meet them. It's worth checking whether that warning exists before adapting the system.
And the fine-tuning case
It deserves a mention because it's the question that comes up most, and it already has its own article covering the full ladder — using, modifying, rebranding. It isn't repeated here.
What this one adds is the framework: fine-tuning doesn't change your role on its own. It changes it if it puts you in one of the three circumstances of Article 25(1), so the useful question isn't "how much have I fine-tuned?" but "under what brand does this go out, what do I say it's for, and is the system high-risk or does it become so?"
The three questions that settle it
Under what brand is it provided? If it's yours, on top of someone else's high-risk system, letter (a) — subject to the express reservation the article itself makes for contractual arrangements.
Have you changed a high-risk system beyond what the manufacturer contemplated in its conformity assessment? And if so, does it affect the requirements or the assessed purpose, and does it remain high-risk?
Have you modified the intended purpose of a system that wasn't high-risk, so that it becomes high-risk? It's the quietest of the three routes, because it leaves no technical trace.
If all three answers are no, you remain the deployer, and the five obligations that open this article don't fall on you, subject to the Article 73 caveat noted at the start. If any is yes, it isn't one that falls on you: they all do.
Content in accordance with Articles 3, 16 and 25 of Regulation (EU) 2024/1689, as amended by Regulation (EU) 2026/1744 (Official Journal of the EU, 24 July 2026).
This article is for informational purposes only and does not constitute legal advice.