Article 2(12) provides that the AI Act does not apply to AI systems released under free and open source licences, unless they are placed on the market or put into service as high risk AI systems or as AI systems falling under Article 5 or Article 50. That is a scope exclusion for AI systems, not a general open source exemption. It does not apply in the same way to providers of general purpose AI models, whose relief is in Articles 53(2) and 54(6) and is dealt with in the next guide. A system that is released openly can therefore still be in scope where it is high risk, prohibited or subject to transparency duties, and a downstream user who places it on the market or puts it into service for a high risk use can acquire provider duties under Article 25.
What does Article 2(12) actually say?
Article 2(12) states, in substance, that the Regulation does not apply to AI systems released under free and open source licences, unless they are placed on the market or put into service as high risk AI systems or as AI systems falling under Article 5 or Article 50. Three elements must be examined. First, the release must be under a free and open source licence. The Act does not define that licence in full in Article 2(12), and the licence terms, including any restrictions on use, modification, distribution or commercial use, must be examined. Second, the system must be released, which raises questions about what counts as release, who releases it and whether making code available in a repository, without more, is a release for this purpose. Third, the exclusion is lost where the system is placed on the market or put into service as a high risk system or as a system under Article 5 or Article 50. The exclusion therefore depends on the licence, the release and the subsequent placing or use. A record that states a system is open source without examining those three elements does not show that Article 2(12) applies. The licence, the release evidence and the placing or use evidence should all be recorded.
What are the three limits in Article 2(12)?
The three limits in Article 2(12) are high risk, prohibited practices and transparency. Where a system released under an open source licence is placed on the market or put into service as a high risk AI system under Article 6, the Act applies to it, and the provider and deployer duties for high risk systems apply. Where the system falls under Article 5, the prohibition applies, and the open source release does not make a prohibited practice lawful. Where the system falls under Article 50, the transparency duties apply, including duties to inform users that they are interacting with an AI system, to mark synthetic content and to disclose the use of emotion recognition or biometric categorisation where the Article requires it. Those limits must be screened in order: prohibition first, then classification, then transparency. A system that passes the prohibition screen can still be high risk, and a system that is not high risk can still be subject to transparency duties. The screening should be recorded for each release, with the version and the licence relied on, because a later release under a different licence, or a later use for a different purpose, can change the answer.
For the broader regulatory framework, see SMEs, startups and open source overview.
What about components, tools and downstream use?
Article 2(12) addresses AI systems, and the Act contains related provisions for third parties that make tools, services, processes or components available. Article 25(4) provides, in substance, that a third party that supplies tools, services, processes or components, other than general purpose AI models, under a free and open source licence is not considered to be the provider of a high risk system on that ground alone, subject to the conditions in the Article. That provision protects component suppliers in defined circumstances, but it does not protect a person who places a high risk system on the market, who substantially modifies a system, or who changes its intended purpose so that it becomes high risk. Downstream use therefore matters. A component released openly can be integrated by a downstream provider into a high risk system, and the downstream provider will carry the provider duties for that system. The downstream provider should obtain the information it needs from the component supplier, including limitations and failure modes, through the documentation and the licence terms, and should record the integration, the testing and the classification. A component supplier should state the intended purpose and the limitations clearly, and should not present a component as suitable for a high risk use where it has not been assessed for that use.
How should a licence be examined for this purpose?
A licence should be examined for whether it is a free and open source licence in the sense used by the Act, whether it allows access, usage, modification and distribution, whether it imposes restrictions that are relevant to the Act, and whether it is the licence actually applied to the release. A licence that restricts use to research only, that caps the number of users, that requires a separate commercial licence for commercial use, or that withholds the parameters, weights or usage information needed to use the system, may not meet the test, depending on the provision relied on and the guidance that applies. Monetisation also matters. Recital 103 states, in substance, that AI components provided against a price or otherwise monetised should not benefit from the open source exceptions, while making components available through open repositories should not in itself constitute monetisation. Recitals do not create obligations, but they guide interpretation, and the monetisation question should be examined with the licence and the commercial arrangements together. The licence examination should be recorded by a person competent to read licences, with the licence text, the version and the date recorded. A licence label, such as open source, in a repository description is not a substitute for the licence text.
What should a downstream deployer check?
A downstream deployer that uses an openly released system should check the licence, the release, the intended purpose, the classification screening for its own use, the Article 5 screen, the Article 50 transparency duties, the data protection position, and the information available about limitations, failure modes and updates. It should check whether its use is within the intended purpose stated by the provider, or whether it changes the intended purpose so that the system becomes high risk under Article 25(1)(c). It should check whether it substantially modifies the system, for example by retraining it on new data, by integrating it with other systems in a way that changes its behaviour, or by putting its own name on it. It should record the version used, the licence, the screening and the oversight arrangements, and should plan for updates and for the withdrawal of the upstream release. A deployer that uses an open source system without that record will struggle to show that it met its deployer duties, and will struggle to respond if the upstream project changes, is withdrawn or is found to have a safety issue. The record should be kept even where the system is free to use, because the absence of a price does not remove the duties.
Continue with open source general purpose models.
What does this guide not decide?
This guide does not decide whether any named licence is a free and open source licence for AI Act purposes, does not decide whether any named release is excluded by Article 2(12), and does not decide whether any downstream use is high risk, prohibited or subject to transparency duties. Licence interpretation, release evidence and classification all need the specific documents and facts. A developer or deployer with a specific release or use should confirm the licence position, the Article 2(12) position and the downstream duties with a competent authority or legal adviser, and should not treat a repository label or a general statement about open source as a legal determination.
Frequently asked questions
Does releasing a system under an open licence remove all AI Act duties?
No. Article 2(12) excludes AI systems released under free and open source licences except where they are placed on the market or put into service as high risk systems or as systems under Article 5 or Article 50. The licence, the release and the subsequent placing or use must all be examined, and downstream providers and deployers can acquire duties for their own use.
What if a system is open source but is used for a high risk purpose?
The Act applies to the high risk use. The person who places the system on the market or puts it into service for that use, and the deployer who uses it, must meet the high risk duties for that use. The open source release does not make the high risk use lawful without those duties, and a change of intended purpose can make a downstream person the provider under Article 25.
Does a research only licence count as open source for Article 2(12)?
That must be examined against the licence text and the provision relied on. A licence that restricts use, caps users or requires a separate commercial licence may not meet the test for the relief in question. The licence text, not the label, should be recorded and confirmed with a legal adviser where the answer matters.
Are open source components protected when they are integrated into a high risk system?
Article 25(4) gives defined protection to third parties that supply tools, services, processes or components, other than general purpose models, under a free and open source licence, but it does not protect a person who places a high risk system on the market, substantially modifies it or changes its intended purpose. The downstream provider carries the provider duties for the integrated system, and the integration and testing must be recorded.
Continue through this AI Act cluster
This article is an informational overview, not a legal determination. Check the current legal text and official implementation guidance for your system and role.
Consolidated Regulation (EU) 2024/1689 ↗AI Act Service Desk timeline ↗