# Creative Commons licenses

> Source: https://aiwiki.ai/wiki/creative_commons_licenses
> Updated: 2026-11-17
> Fact-checked: 2026-11-17
> Categories: AI Policy & Regulation, Open Source AI
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Creative Commons licenses." aiwiki.ai, 17 Nov 2026. https://aiwiki.ai/wiki/creative_commons_licenses
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

**Creative Commons licenses** are a set of standardized public copyright licenses, published by the nonprofit organization Creative Commons, that let a creator grant the public permission to copy, share and in most cases adapt a work under a small number of conditions. The current suite, version 4.0, consists of six licenses built from four elements (Attribution, ShareAlike, NonCommercial and NoDerivatives), alongside the separate CC0 public domain dedication.[1][2][3] The licenses were written for creative works such as text, images, music and video, but in [artificial intelligence](https://aiwiki.ai/wiki/artificial_intelligence) they turn up in two places. Developers release trained parameters ([open weights](https://aiwiki.ai/wiki/open_weights)) under them, as with AlphaFold 2 (CC BY 4.0) and MusicGen (CC BY-NC 4.0).[4][5] In early 2024 Creative Commons licenses were the fourth most common license family on [Hugging Face](https://aiwiki.ai/wiki/hugging_face) model repositories.[6] They also cover many [datasets](https://aiwiki.ai/wiki/datasets) used for training, including Wikipedia text, Mozilla's Common Voice recordings and, from 2022 to 2023, the metadata of the LAION-5B image index.[7][8][9]

Both uses raise questions the licenses were not designed for. Creative Commons recommends against using its licenses for software, and it has said repeatedly that its licenses cannot stop uses of a work, such as AI training, that copyright law already permits without a license.[10][11] Legal scholars have questioned whether model weights are protected by copyright at all, which matters because a CC license only operates where copyright or a similar right exists.[10][12] A license that limits use to noncommercial purposes also keeps a model outside the [Open Source AI Definition](https://aiwiki.ai/wiki/open_source_ai_definition) and, under European Commission guidelines, outside the [EU AI Act](https://aiwiki.ai/wiki/eu_ai_act)'s exemptions for free and open-source models.[13][14] Since 2025 Creative Commons has been developing "CC signals", a framework meant to set terms for AI reuse of shared collections, which as of November 2026 was heading for a public alpha test in 2027.[15][16]

## Background

Lawrence Lessig, Hal Abelson and Eric Eldred created Creative Commons in 2001 and published the first licenses in 2002. The organization launched CC0 in 2009.[1] Version 4.0 of the licenses was published on November 25, 2013. Creative Commons described the 4.0 suite as available for adoption worldwide, and it extended the licenses to cover sui generis database rights, which exist in the European Union and a few other countries.[2]

The four elements are defined by Creative Commons as follows:[3]

- **BY (Attribution):** credit must be given to the creator.
- **SA (ShareAlike):** adaptations must be shared under the same terms.
- **NC (NonCommercial):** only noncommercial use of the work is permitted.
- **ND (NoDerivatives):** no derivatives or adaptations of the work are permitted.

Every one of the six licenses includes Attribution.[3] CC0 is not a license in the same sense: it lets a rights holder waive copyright and related rights "to the fullest extent permitted by law", placing the work as fully as possible into the public domain.[17]

| Tool | Elements | Commercial use | Adaptations may be shared | Condition on shared adaptations |
| --- | --- | --- | --- | --- |
| CC0 1.0 | None (waiver) | Yes | Yes | None |
| CC BY 4.0 | BY | Yes | Yes | Credit |
| CC BY-SA 4.0 | BY, SA | Yes | Yes | Credit; same or compatible license |
| CC BY-ND 4.0 | BY, ND | Yes | No | Not applicable |
| CC BY-NC 4.0 | BY, NC | No | Yes | Credit; noncommercial only |
| CC BY-NC-SA 4.0 | BY, NC, SA | No | Yes | Credit; noncommercial; same or compatible license |
| CC BY-NC-ND 4.0 | BY, NC, ND | No | No | Not applicable |

Several features of the 4.0 legal code bear directly on AI use. The license applies only to "Copyright and Similar Rights" the licensor holds; where an exception or limitation to copyright covers a use, the license "does not apply". Patent and trademark rights are not licensed. A licensee who breaches the terms loses the license automatically but gets it back if the violation is cured within 30 days of discovering it.[18] The Creative Commons FAQ adds that the licenses "are operative only when applied to material in which a copyright exists".[10]

### What "NonCommercial" means

The 4.0 legal code defines NonCommercial as "not primarily intended for or directed towards commercial advantage or monetary compensation."[18] Creative Commons says the test turns on the use, not on the user: a nonprofit can breach the NC term, and a for-profit company does not necessarily breach it by using an NC-licensed work. "Whether a use is commercial will depend on the specifics of the situation and the intentions of the user," the FAQ says, and Creative Commons declines to advise on particular cases, telling uncertain users to ask the rights holder.[10] The FAQ addresses one AI-adjacent case: a for-profit company that uses NC-licensed content and releases the result under research-only terms. Whether that breaches the NC term, it answers, depends on whether any of the company's uses are primarily intended for commercial advantage, which "is hard to know without having all of the facts".[10]

## Model weights under Creative Commons licenses

Model developers who use CC licenses usually pair a software license for the code with a CC license for the weights, as the examples below show. A study of the Hugging Face Hub published in February 2024 counted 174,163 model repositories that declared a license; 7.5% of them used a Creative Commons license, behind Apache (38.1%), the [OpenRAIL](https://aiwiki.ai/wiki/openrail) and other RAIL licenses (24.0%) and MIT (17.5%). Ranked by downloads rather than by repository count, Creative Commons licenses came third, ahead of RAIL.[6] Hugging Face's license metadata lists identifiers for CC0 and for CC licenses back to version 2.0, including `cc-by-4.0`, `cc-by-nc-4.0` and `cc-by-nc-sa-4.0`.[19]

| Model | Developer | Weights license | Code license | Notes |
| --- | --- | --- | --- | --- |
| [AlphaFold](https://aiwiki.ai/wiki/alphafold) 2 parameters | DeepMind | CC BY 4.0 (CC BY-NC 4.0 until January 19, 2022) | Apache 2.0 | Relicensed to permit commercial use[20][4] |
| [GenCast](https://aiwiki.ai/wiki/gencast) | Google DeepMind | CC BY 4.0 (previously CC BY-NC-SA 4.0) | Apache 2.0 | Weights relicensed August 6, 2026[21] |
| NLLB-200 ([No Language Left Behind](https://aiwiki.ai/wiki/no_language_left_behind)) | Meta | CC BY-NC 4.0 | MIT | Translation models for 200 languages[22] |
| [Massively Multilingual Speech](https://aiwiki.ai/wiki/massively_multilingual_speech) | Meta | CC BY-NC 4.0 | CC BY-NC 4.0 | Code and weights under the same CC license[23] |
| [MusicGen](https://aiwiki.ai/wiki/musicgen) (AudioCraft) | Meta | CC BY-NC 4.0 | MIT | Text-to-music models[5] |
| [DINOv2](https://aiwiki.ai/wiki/dinov2) | Meta | Apache 2.0 (CC BY-NC until August 31, 2023) | Apache 2.0 (CC BY-NC until August 31, 2023) | Code and models relicensed together[24] |
| [Parakeet](https://aiwiki.ai/wiki/parakeet) TDT 0.6B v2 | [NVIDIA](https://aiwiki.ai/wiki/nvidia) | CC BY 4.0 | Not listed on the card | Speech recognition[25] |
| [Command A](https://aiwiki.ai/wiki/command_a) | [Cohere](https://aiwiki.ai/wiki/cohere) Labs | CC BY-NC 4.0 plus Cohere Labs' Acceptable Use Policy | Not listed on the card | 111 billion parameters[26] |
| [NVLM](https://aiwiki.ai/wiki/nvlm)-D-72B | NVIDIA | CC BY-NC 4.0 | Not listed on the card | Card also points to the Qwen2 and MIT licenses of its components[27] |

### Relicensing from NonCommercial to Attribution

Several developers have released weights under a noncommercial CC license and later dropped the NC element. DeepMind's AlphaFold 2 repository originally stated that the parameters were "made available for non-commercial use only" under CC BY-NC 4.0. A commit dated January 19, 2022, titled "Update license of AlphaFold parameters", replaced that wording with CC BY 4.0 and pointed the download script and the Colab notebook at new parameter archives dated 2022-01-19.[20] The same commit changed the text of the AlphaFold Colab notebook, which had said that because it used the parameters, "its outputs are thus for non-commercial use only"; the new text no longer made any claim about outputs.[20] As of November 2026 the AlphaFold 2 README still lists the parameters under CC BY 4.0.[4]

GenCast followed the same path more than four years later. Its README in the WeatherNext repository carries a notice dated August 6, 2026 saying that "the license for the model weights in this repository has been updated to permit commercial use", and the weights are now under CC BY 4.0.[21] Meta changed the code and model license of DINOv2 from CC BY-NC to Apache 2.0 in a commit on August 31, 2023.[24]

CC licenses have also been used for code, against Creative Commons' own advice. When AlphaFold 3 was released in November 2024 its inference code was under CC BY-NC-SA 4.0; a commit on June 9, 2026 relicensed the code to Apache 2.0, stating that "only the code is re-licensed, the weights license and terms of use stay the same." The AlphaFold 3 parameters are distributed under Google's own AlphaFold 3 Model Parameters Terms of Use rather than a CC license.[28][29]

### Creative Commons' advice on software

The Creative Commons FAQ says: "We recommend against using Creative Commons licenses for software." It gives three reasons: CC licenses contain no terms about distributing source code, they do not address patent rights, and they are not compatible with the major software licenses, which makes it hard to combine CC-licensed material with other free software. It recommends licenses that the Free Software Foundation lists as free and the [Open Source Initiative](https://aiwiki.ai/wiki/open_source_initiative) lists as open source. CC0 is the exception: Creative Commons describes it as GPL-compatible and acceptable for software. CC BY-SA 4.0 is one-way compatible with the GNU General Public License version 3.[10]

The FAQ does not mention model weights, and whether its reasoning carries over depends on whether weights are treated as software. Peter Henderson of Princeton University and Mark Lemley of Stanford Law School argue that weights, unlike object code, are not derived from code written by a human.[12] The Linux Foundation's [Model Openness Framework](https://aiwiki.ai/wiki/model_openness_framework) specification treats parameters as "data for which an open source software license like Apache 2.0 and MIT are not well suited" and recommends an open data license such as CDLA-Permissive-2.0.[30] The patent point applies to weights in any case. The [Apache License 2.0](https://aiwiki.ai/wiki/apache_license) includes a patent license from each contributor, while the CC 4.0 legal code states that patent rights are not licensed, so a model released under CC BY 4.0 carries no patent permission from the licensor, whereas one released under Apache 2.0 does.[18][31] In four of the examples above (AlphaFold 2, GenCast, NLLB-200 and MusicGen), CC-licensed weights and Apache 2.0 or [MIT License](https://aiwiki.ai/wiki/mit_license) code are released as separate components, each under its own terms; Meta's MMS put code and weights under the same CC BY-NC 4.0 license.

## Training datasets

Creative Commons licensing is common in the data used to train and evaluate models.

| Dataset | Terms | Notes |
| --- | --- | --- |
| Common Voice (Mozilla) | CC0 | Voice recordings and sentences dedicated to the public domain; 18 languages and about 1,400 hours at the February 2019 release[8] |
| Wikipedia and other Wikimedia text | CC BY-SA 4.0 and the GNU Free Documentation License | Licenses contributors grant under the terms of use in effect since June 7, 2023; reusers may follow either[7] |
| [LAION](https://aiwiki.ai/wiki/laion)-5B (2022 release) | CC BY 4.0 for the metadata | 5.85 billion image-text pairs; the images themselves remain "under their copyright". Withdrawn in December 2023; the cleaned Re-LAION-5B (August 2024) is under Apache 2.0[9][47] |
| BFD (as mirrored for AlphaFold) | CC BY-SA 4.0 | Protein sequence database[4] |
| MGnify protein database (as mirrored for AlphaFold) | CC0 1.0 | Protein sequence database[4] |
| [Common Pile](https://aiwiki.ai/wiki/common_pile) v0.1 | Mix of CC0, CC BY, CC BY-SA, public domain and other open licenses | 8 TB from 30 sources; NC and ND content excluded[32] |

Common Voice's distribution changed in 2025. A Mozilla post of July 30, 2025 said that existing Common Voice datasets "will continue to be accessible under the CC-0 licence they were released with" and that future versions would be released through Mozilla Data Collective, a platform that lets dataset owners choose their own terms.[33] As of September 2026 the scripted-speech dataset was at version 27.0 and covered 295 languages.[34]

LAION's original licensing illustrates a common split: it licensed the LAION-5B metadata files under CC BY 4.0 and made no license claim over the underlying images.[9] LAION withdrew LAION-5B in December 2023 after the Stanford Internet Observatory found links to child sexual abuse material, and released the cleaned Re-LAION-5B under Apache 2.0 in August 2024.[47] The Common Pile, released by [EleutherAI](https://aiwiki.ai/wiki/eleutherai) and collaborators in June 2025, took the opposite approach and included only text it could trace to an open license. Its paper adopts the Open Knowledge Foundation's Open Definition 2.1 and states that "CC NC (non-commercial) and ND (no derivatives) licenses are not considered open under this definition"; it also warns against "license laundering", in which a work is redistributed, typically by someone who does not hold the rights, with an incorrect license.[32]

### Attribution and ShareAlike in training

Whether training on CC-licensed works triggers the license conditions depends first on whether training needs copyright permission at all. The Creative Commons FAQ states that "CC licenses never limit uses that copyright doesn't control", and that "as a general matter text and data mining in the United States is considered a fair use", so reusers doing text and data mining there do not have to meet the attribution requirements, although Creative Commons "strongly recommend[s]" that they do.[10] Where copyright is engaged and a CC-licensed work is distributed inside a dataset, attribution must be given in a way that is reasonable in the circumstances, which may be as simple as a link to a page carrying the attribution information.[10]

Creative Commons' guide to using CC-licensed works for AI training gives conservative advice for developers who want to follow each element even where copyright may not require it, and warns that this approach "is likely to lead to overcompliance". Its legal analysis is narrower: the BY, SA and ND conditions are "triggered only when works or adaptations of works are publicly shared", while NC "applies to all uses requiring permission under copyright".[11]

| Element | Creative Commons' conservative guidance |
| --- | --- |
| BY | All CC licenses require attribution; for training, attribution "could be a simple link to the source of the dataset used to train the model" |
| SA | If models or outputs are based on ShareAlike content and are shared publicly, the developer would have to use the same CC license as the original works |
| NC | Applies to every use requiring permission: "all stages, from copying the data during training to sharing the trained model, must not be for commercial gain" |
| ND | Following the restriction "would require that ND-licensed content not be used as training data" |

For datasets, the FAQ says ShareAlike applies only when a work is modified and shared publicly. A published dataset of SA-licensed photos is likely a collection, which does not itself have to carry an SA license, although each photo remains under its own license.[10] The harder question is whether a trained model or its output is an adaptation of the training works at all. Andrés Guadamuz, a reader in intellectual property law at the University of Sussex, argued in 2023 that training generally copies works without sharing them and extracts statistical information, so it creates neither a shared copy nor an adaptation and does not trigger the attribution, NonCommercial or ShareAlike conditions; he noted that the United States fair use question was still being litigated.[35] AI Wiki's article on [copyright and AI](https://aiwiki.ai/wiki/copyright) covers the court cases and national exceptions in detail.

Rights holders have also asked for attribution outside the license framework. On November 10, 2025 the Wikimedia Foundation, whose projects' text is under CC BY-SA, asked that generative AI give "credit to the human contributions that it uses to create its outputs" and said that most AI developers "should properly access Wikipedia's content through the Wikimedia Enterprise platform", its paid access service.[36]

## Creative Commons' positions on AI

Creative Commons has consistently said that its licenses cannot be used to forbid AI uses that the law already allows. In December 2021 it stated that CC license terms "cannot be construed or interpreted as a reservation of a right" under Article 4 of the EU Directive on Copyright in the Digital Single Market, the text and data mining exception that rights holders can opt out of, and that "CC licenses do not operate as an opt-out of an exception or limitation to copyright."[37] In an August 2023 post, Creative Commons wrote that "there are strong arguments that, in most cases, using copyrighted works to train generative AI models would be fair use in the United States", and that "you cannot use the licenses to prohibit a use if it is otherwise permitted by limitations and exceptions."[38]

A legal primer published in May 2025 repeated that "AI training is often permitted by copyright", concluded that "the CC license conditions have limited application to machine reuse", and advised that "using a more restrictive CC license in an effort to prevent AI training is not an effective approach."[39] In September 2026 Creative Commons published licensing guidance for an AI ecosystem, and in October it held two virtual events on how the guidance applies to education, science and culture. It said its long-standing advice to share under the most flexible license or tool that fits a licensor's goals "hasn't substantially changed", that "licensing alone cannot address all of the challenges to sharing that AI presents", and that CC signals "does not replace or modify CC licenses."[40]

### CC signals

Creative Commons announced the public kickoff of CC signals on June 25, 2025, describing it as "a new preference signals framework designed to increase reciprocity and sustain a creative commons in the age of AI." It said the signals "may range in enforceability, legally binding in some cases and normative in others", and it aimed for an alpha launch in November 2025.[15] The draft proposal on GitHub was meant for stewards of large content collections and tied the signals to the vocabulary for AI usage preferences then being developed at the Internet Engineering Task Force. It offered four proposed signals: Credit, Credit plus Direct Contribution, Credit plus Ecosystem Contribution, and Credit plus Open.[41]

The design changed after public consultation. In April 2026 Creative Commons wrote that feedback had shown that "preference signals without enforcement do not meaningfully shift power" and that CC signals was no longer limited to signaling preferences.[42] A May 13, 2026 post by Anna Tumadóttir and Sarah Hinchliff Pearson described a shift "from expressing preferences to rebalancing power to protect the commons". It said that attribution, treated as a preference in the first version, "must be a requirement", announced work on a tool that "will allow data stewards to set terms for accessing and using a collection or compilation", with heavier conditions for resource-intensive bulk reusers and public interest uses excluded, and set a goal of raising $5 million for the next iteration.[43]

As of November 2026, CC signals was a contract-based legal tool still in development. After a convening in Chicago on November 2 and 3, 2026, Creative Commons said on November 16 that it was refining the first legal tool for a public alpha in early 2027. The tool is a standardized contract for organizations offering bulk access to their collections, through an API or a dataset download. The alpha will focus on "SourceConnection", under which participating AI developers would disclose which collections they trained on and provide references when systems draw on outside sources; "ShareBack" requirements, such as sharing access metrics, modified datasets or money, "remain under consideration for future iterations."[16]

The project has drawn criticism. Writing in The Scholarly Kitchen in May 2026, Rick Anderson argued that the signals "ultimately constitute requests, not legally-binding criteria", that an AI opt-out cannot narrow a CC BY license already granted because those licenses are irrevocable, and that terms such as "public interest" and "responsible" reuse were undefined.[44]

## Are model weights copyrightable?

Because a CC license operates only where copyright or a similar right exists, the legal status of weights decides whether a CC license on them binds anyone.[10] Commentators describe the question as unsettled. Henderson and Lemley, in a 2024 article on the enforceability of AI model terms, argue that weights "are also created through an automated process, likely rendering them uncopyrightable", and that "even if the human authorship requirement is met, model weights are likely functional artifacts unprotected by copyright." They consider and reject the argument that a model is a protected database or compilation of its training data as "an uphill battle".[12] They conclude that if weights are not copyrightable, open-weight models may be "more 'open' than they would seem at a glance", because developers would lack an effective means of controlling downstream use.[12] Michael Dolan, senior vice president of legal and strategic programs at the [Linux Foundation](https://aiwiki.ai/wiki/linux_foundation), told TechTarget in June 2026 that a weights file is "basically data at that point. There's not copyrightable works in there."[45]

The Open Source Initiative takes no position on the question: the Open Source AI Definition says it "does not require a specific legal mechanism" for keeping parameters free, and that parameters "may be free by their nature or a license or other legal instrument may be required to ensure their freedom."[13]

## Open source definitions and the EU AI Act

The main definitions of open source exclude licenses that restrict commercial use. The Open Source Definition, which OSI applies to software licenses, requires that a license "must not restrict anyone from making use of the program in a specific field of endeavor", and gives use in a business as an example of a field that may not be excluded.[46] The Open Source AI Definition 1.0 requires the freedom to "use the system for any purpose and without having to ask for permission", and requires that parameters be made available under "OSI-approved terms".[13] A license limited to noncommercial use, such as the CC BY-NC 4.0 license on NLLB-200 or MusicGen, restricts the kind of field of use these texts rule out, and the Common Pile authors reached the same conclusion for the Open Definition, under which "CC NC (non-commercial) and ND (no derivatives) licenses are not considered open".[32] AI Wiki's [list of open-weight models](https://aiwiki.ai/wiki/list_of_open_weight_models) marks CC BY-NC weights as not licensed for commercial use, and its [comparison of open-weight licenses](https://aiwiki.ai/wiki/open_weight_license_comparison) groups non-commercial licenses in its third tier, "conditional or non-commercial".

The European Commission's guidelines on general-purpose AI models under the AI Act interpret the exemption for models released under a "free and open-source licence". They state that the license should allow use "without any restrictions besides restrictions ensuring attribution and distribution of the model or derivatives under the same or comparable terms", that distribution may be subject to limited conditions such as crediting the authors, and that "limitations to non-commercial or research-only use" are among the restrictions that disqualify a license.[14] The guidelines therefore accept Attribution and ShareAlike-style conditions and exclude NonCommercial ones. They do not name Creative Commons licenses. The exemption these guidelines interpret, in Articles 53(2) and 54(6), covers only general-purpose AI models, does not apply to a model with systemic risk, and does not lift the duties to keep a copyright policy and to publish a summary of training content. The guidelines also require that the model is not monetised and allow proportionate, safety-oriented use restrictions.[14]

## See also

- [Open weights](https://aiwiki.ai/wiki/open_weights)
- [Open-weight LLM license comparison](https://aiwiki.ai/wiki/open_weight_license_comparison)
- [OpenRAIL](https://aiwiki.ai/wiki/openrail)
- [OpenMDW](https://aiwiki.ai/wiki/openmdw)
- [Open Source AI Definition](https://aiwiki.ai/wiki/open_source_ai_definition)
- [Common Corpus](https://aiwiki.ai/wiki/common_corpus)
- [Fair use](https://aiwiki.ai/wiki/fair_use)

## References

1. Creative Commons. "History". creativecommons.org. https://creativecommons.org/history/
2. Diane Peters. "CC's Next Generation Licenses: Welcome Version 4.0!". Creative Commons, November 25, 2013. https://creativecommons.org/2013/11/25/ccs-next-generation-licenses-welcome-version-4-0/
3. Creative Commons. "About CC Licenses". creativecommons.org. https://creativecommons.org/share-your-work/cclicenses/
4. Google DeepMind. "AlphaFold" README, "Model Parameters License" and "Mirrored Databases". GitHub, google-deepmind/alphafold. https://github.com/google-deepmind/alphafold
5. Meta AI. "AudioCraft" README, "License". GitHub, facebookresearch/audiocraft. https://github.com/facebookresearch/audiocraft
6. Daniel McDuff, Tim Korjakow, Scott Cambo, Jesse Josua Benjamin, Jenny Lee, Yacine Jernite, Carlos Muñoz Ferrandis, Aaron Gokaslan, Alek Tarkowski, Joseph Lindley, A. Feder Cooper and Danish Contractor. "On the Standardization of Behavioral Use Clauses and Their Adoption for Responsible Licensing of AI". arXiv:2402.05979, February 2024. https://arxiv.org/abs/2402.05979
7. Wikimedia Foundation. "Terms of Use", section 7. foundation.wikimedia.org, effective June 7, 2023. https://foundation.wikimedia.org/wiki/Policy:Terms_of_Use
8. Mozilla. "Sharing our Common Voices: Mozilla releases the largest to-date public domain transcribed voice dataset". The Mozilla Blog, February 28, 2019. https://blog.mozilla.org/en/mozilla/news/sharing-our-common-voices-mozilla-releases-the-largest-to-date-public-domain-transcribed-voice-dataset/
9. LAION. "LAION-5B: A New Era of Open Large-Scale Multi-Modal Datasets". laion.ai, March 31, 2022. https://laion.ai/blog/laion-5b/
10. Creative Commons. "Frequently Asked Questions". creativecommons.org. https://creativecommons.org/faq/
11. Creative Commons. "Using CC-Licensed Works for AI Training". creativecommons.org. https://creativecommons.org/using-cc-licensed-works-for-ai-training-2/
12. Peter Henderson and Mark A. Lemley. "The Mirage of Artificial Intelligence Terms of Use Restrictions". arXiv:2412.07066, December 2024 (forthcoming, Indiana Law Journal). https://arxiv.org/abs/2412.07066
13. Open Source Initiative. "The Open Source AI Definition 1.0". opensource.org. https://opensource.org/ai/open-source-ai-definition
14. European Commission. "Commission Guidelines on the scope of the obligations for providers of general-purpose AI models established by Regulation (EU) 2024/1689 (AI Act)", C(2025) 7719 final, November 19, 2025 (content approved July 18, 2025), paragraphs 69-81. digital-strategy.ec.europa.eu. https://digital-strategy.ec.europa.eu/en/library/guidelines-scope-obligations-providers-general-purpose-ai-models-under-ai-act
15. Creative Commons. "Introducing CC Signals: A New Social Contract for the Age of AI". Creative Commons, June 25, 2025. https://creativecommons.org/2025/06/25/introducing-cc-signals-a-new-social-contract-for-the-age-of-ai/
16. Creative Commons. "Advancing CC Signals: Preparing for Public Alpha Testing in 2027". Creative Commons, November 16, 2026. https://creativecommons.org/2026/11/16/advancing-cc-signals-preparing-for-public-alpha-testing-in-2027/
17. Creative Commons. "CC0". creativecommons.org. https://creativecommons.org/public-domain/cc0/
18. Creative Commons. "Attribution-NonCommercial 4.0 International, Legal Code". creativecommons.org. https://creativecommons.org/licenses/by-nc/4.0/legalcode.en
19. Hugging Face. "Licenses". Hugging Face Hub documentation. https://huggingface.co/docs/hub/repositories-licenses
20. Google DeepMind. "Update license of AlphaFold parameters" (commit 8173117). GitHub, google-deepmind/alphafold, January 19, 2022. https://github.com/google-deepmind/alphafold/commit/8173117130e6df8749ab7349722ec465666df548
21. Google DeepMind. "WeatherNext 1 Gen (GenCast)" README, "License and Disclaimers". GitHub, google-deepmind/weathernext (formerly google-deepmind/graphcast). https://github.com/google-deepmind/weathernext/blob/main/docs/weathernext1_gen/README.md
22. Meta AI. "No Language Left Behind" README. GitHub, facebookresearch/fairseq (nllb branch). https://github.com/facebookresearch/fairseq/tree/nllb
23. Meta AI. "MMS: Scaling Speech Technology to 1000+ languages" README, "License". GitHub, facebookresearch/fairseq. https://github.com/facebookresearch/fairseq/tree/main/examples/mms
24. Meta AI. "Update license everywhere (#182)" (commit 81b2b64). GitHub, facebookresearch/dinov2, August 31, 2023. https://github.com/facebookresearch/dinov2/commit/81b2b6419385a321287de91e00282ef7cbd26f94
25. NVIDIA. "parakeet-tdt-0.6b-v2" model card, "License/Terms of Use". Hugging Face. https://huggingface.co/nvidia/parakeet-tdt-0.6b-v2
26. Cohere Labs. "c4ai-command-a-03-2025" model card. Hugging Face. https://huggingface.co/CohereLabs/c4ai-command-a-03-2025
27. NVIDIA. "NVLM-D-72B" model card, "License/Terms of Use". Hugging Face. https://huggingface.co/nvidia/NVLM-D-72B
28. Google DeepMind. "Change the code license from CC BY-NC-SA 4.0 to Apache 2.0" (commit 7b197fe). GitHub, google-deepmind/alphafold3, June 9, 2026. https://github.com/google-deepmind/alphafold3/commit/7b197fe859790fc3e04d03ea70dd0b9ba48881c9
29. Google DeepMind. "AlphaFold 3" README, "Licence and Disclaimer". GitHub, google-deepmind/alphafold3. https://github.com/google-deepmind/alphafold3
30. LF AI & Data Generative AI Commons. "The Model Openness Framework (MOF) Specification", version 1.0. LF AI & Data, December 17, 2024. https://lfaidata.foundation/wp-content/uploads/sites/3/2025/01/05_White_paper_MOF_Specification.pdf
31. Apache Software Foundation. "Apache License, Version 2.0", section 3 ("Grant of Patent License"). apache.org, January 2004. https://www.apache.org/licenses/LICENSE-2.0
32. Nikhil Kandpal, Brian Lester, Colin Raffel et al. "The Common Pile v0.1: An 8TB Dataset of Public Domain and Openly Licensed Text". arXiv:2506.05209, June 5, 2025. https://arxiv.org/abs/2506.05209
33. Francis Tyers. "Exciting news! Mozilla Data Collective". Mozilla Discourse, July 30, 2025. https://discourse.mozilla.org/t/exciting-news-mozilla-data-collective/145191
34. Mozilla. "Common Voice Datasets" README. GitHub, common-voice/cv-dataset. https://github.com/common-voice/cv-dataset
35. Andrés Guadamuz. "Creative Commons and AI training". TechnoLlama, November 19, 2023 (updated December 11, 2023). https://technollama.co.uk/creative-commons-and-ai-training
36. Wikimedia Foundation. "In the AI era, Wikipedia has never been more valuable". Wikimedia Foundation, November 10, 2025. https://wikimediafoundation.org/news/2025/11/10/in-the-ai-era-wikipedia-has-never-been-more-valuable/
37. Brigitte Vézina. "Creative Commons' Statement on CC licenses and the Text and Data Mining Exception Under Article 4 EU CDSM Directive". Creative Commons, December 17, 2021. https://creativecommons.org/2021/12/17/creative-commons-statement-on-cc-licenses-and-the-ext-and-data-mining-exception-under-article-4-eu-cdsm-directive/
38. Kat Walsh. "Understanding CC Licenses and Generative AI". Creative Commons, August 18, 2023. https://creativecommons.org/2023/08/18/understanding-cc-licenses-and-generative-ai/
39. Sarah Hinchliff Pearson. "Understanding CC Licenses and AI Training: A Legal Primer". Creative Commons, May 15, 2025. https://creativecommons.org/2025/05/15/understanding-cc-licenses-and-ai-training-a-legal-primer/
40. Creative Commons. "Guidance on Using CC Licenses in an AI Ecosystem". Creative Commons, September 3, 2026. https://creativecommons.org/2026/09/03/guidance-on-using-cc-licenses-in-an-ai-ecosystem/
41. Creative Commons. "cc-signals" (draft proposal README). GitHub, creativecommons/cc-signals. https://github.com/creativecommons/cc-signals
42. Creative Commons. "Update on CC Signals: What Changed and Why". Creative Commons, April 23, 2026. https://creativecommons.org/2026/04/23/update-on-cc-signals-what-changed-and-why/
43. Anna Tumadóttir and Sarah Hinchliff Pearson. "From Signals to Infrastructure: Strengthening the Commons for the AI Era". Creative Commons, May 13, 2026. https://creativecommons.org/2026/05/13/from-signals-to-infrastructure-strengthening-the-commons-for-the-ai-era/
44. Rick Anderson. "Seven Questions about CC Signals". The Scholarly Kitchen, May 26, 2026. https://scholarlykitchen.sspnet.org/2026/05/26/seven-questions-about-cc-signals/
45. Sean Michael Kerner. "A legal minefield: Open source licensing for AI models". TechTarget, June 9, 2026. https://www.techtarget.com/searchapparchitecture/feature/The-industry-is-trying-to-fix-AI-model-licensings-legal-minefield
46. Open Source Initiative. "The Open Source Definition". opensource.org. https://opensource.org/osd
47. LAION. "Releasing Re-LAION-5B: transparent iteration on LAION-5B with additional safety fixes". laion.ai, August 30, 2024. https://laion.ai/blog/relaion-5b/

