The Case for a National Cultural Data Sovereignty Strategy

cspc logo

Author(s):

Valentine Goddard

AI Impact Alliance

Founder and Executive Director

UNESCO’s CSO in AI Ethics working group on Intellectual Property and Culture

Co-lead

Disclaimer: The French version of this text has been auto-translated and has not been approved by the author.

Creative economy is the fastest growing economic sector globally. This makes art and cultural data a valuable resource at the center of the geoeconomically charged AI race. We should be afraid, or at least very aware, of where that precious data is flowing.

To Fear or Not to Fear, or What to Fear

The topic of AI has never been gloomier. That may actually not be such a bad thing. According to the late Henry Kissinger, the threat that AI could obliterate humanity could be the driving force to identify shared interests and build forms of international restraint and cooperation. The critical question then becomes: what, exactly, should we fear? How might AI “obliterate” humanity exactly? 

That distinction matters, because what we fear determines how we defend ourselves against it. If we understand the threat as primarily technical, we will invest in technical safeguards. If we understand it as fundamentally social, political and economic, we will invest accordingly, in literacy, policy and regulation, economic alliances, institutional partnerships and multilateral cooperation. So, what is the real threat: the uncontrollable, growing technical abilities of AI, or a fragile, timid patchwork of policies and regulations? 

Allow me a quick analogy. Let’s imagine we have frontier model 2026 race cars on the road, but we are in the 1920s when roads weren’t even ready for Ford Model A. Seat belts, stop signs, speed limits, aren’t a thing yet. Not only do we need to make sure that cars operate safely on the roads, but also that cars are regularly supplied with fuel. The Large Language Models (LLMs) that sustain frontier AI models need data, like engines need fuel. Stagnant, biased, incomplete data is like low grade fuel; corrupt data is like contaminated fuel; and blocked access to data is like blocking the Straits of Hormuz and Bab el-Mandeb. 

Making a resource scarce, or unreliable, triggers both economic and military tremors. So perhaps the deeper threat is not simply that AI is becoming more powerful, but that trusted data alliances, governance institutions and international partnerships need to shape that power collectively, especially as we try to grasp the potential of creative economies. The strategic question, then, is not only who builds the most powerful models, but who builds the most trusted data ecosystems around them. 

Multipronged Value Extraction of Cultural Data

First, AI is increasingly embedded throughout artistic and cultural production. As creators adopt these tools, their data can flow into privately owned and/or government-controlled technological infrastructures, while numerous lawsuits challenge the use of artists’ work without consent or remuneration.

Second, AI platforms are now being used to determine whether art is human. Creators’ works are submitted, with or without their consent, to detection platforms that may lead to accusations of using AI, forcing artists (authors, musicians, etc.) to fight for their individual rights.  

Third, cultural works themselves are becoming AI raw material. Reports document a distressing trend of books being bought, digitized and destroyed. Nippon TV News reported on 50 tons of books shipped to the U.S. and suspected to train proprietary AI systems, echoing practices already documented among U.S. AI companies. Courts have so far found that purchased books can be destroyed, illustrating how ill-equipped our legal system is to protect cultural data. 

Weaponization of Identities

While the deliberate destruction of books has long been a tool of political control, today’s technologically mediated forms of cultural data extraction also contribute to training weapons. Governments and political actors are increasingly interested in AI-generated personas, synthetic media and other tools capable of shaping information environments and influencing public opinion. Anthropic for example has partnerships with Palantir and other U.S. national-security actors to make Claude available in classified defence and intelligence environments. 

Suffice to say that our cultural data is becoming part of a much larger information ecosystem where it is being used to analyse populations, shape narratives, influence political opinion, intensify social divisions, or support security and military applications. The cultural sector’s reliance on AI systems is increasing (faster French to English translation, grant applications, project summaries, production, marketing, etc.), and we are feeding into these data currents.  

A Historical Opportunity

It is urgent and critical to ask ourselves where our ideas, our stories are going. 

“Every institution, library, foundation, cultural group, and government around the world that possesses cultural content should make it available for training, free and open, AI foundation models. Free and open AI systems will constitute the repository of all human knowledge and culture.” Yann LeCun, Chief AI Scientist, Meta. 

This should send chills down our spines. The value extraction of cultural heritage and production data risks collapsing both the Canadian economy and the pluralistic, rights-based, democratic values that make us who we are. Policy changes and data sharing alliances are necessary to grasp the potential of innovation in creative economies, to ensure that the value created from Canadian culture flows back into the communities and creative ecosystems that produced it. We need a national cultural data strategy to protect and retain the value of our language, our stories, our imagined futures.

That is why, in support of UNESCO, AI Impact Alliance brought together experts and cultural organizations, with notable Indigenous and Global South representation, to reimagine institutions and democratic infrastructures for cultural data governance. The results show that cultural sovereignty, consent and fair value-sharing are competitive advantages in creative economies. We invite you to read more about this work to protect cultural data commons and the building of democratic infrastructures in creative economies on our website or on UNESCO’s. 

A national cultural data sovereignty strategy would allow Canada to compete through trust: protecting the value of our culture while building the alliances and institutions needed to shape the next creative economy. 

Author’s Use of AI to Assist the Writing of Article

Ideas and writing style: 100% my own. ChatGPT and Perplexity were used for some information retrieval and editing. In doing so, I acknowledge that my ideas were copied into a prompt, and that prompt feeds into foreign owned data-gathering AI systems, systems that are now being treated as critical national assets and the access to them being restricted by governments. This begs the question, for how much longer will we be allowed to contribute freely to foreign information systems? 

1 https://www.henryakissinger.com/articles/the-path-to-ai-arms-control/
2 https://www.lapresse.ca/arts/arts-visuels/2025-10-15/intelligence-artificielle/des-artistes-visuels-font-une-demande-d-action-collective-contre-meta.php;https://www.lapresse.ca/arts/litterature/2025-09-08/intelligence-artificielle/des-auteurs-quebecois-deposent-une-demande-d-action-collective-contre-microsoft.php
3 Haitian-Quebec-Canadian author, Thélyson Orélien’s was recently accused of using AI to write C’était ça ou mourir. Pangram was used to detect whether he used AI to write the award-winning book, launching a flurry of media coverage on the technical viability and relevance of such tools. Mon entrevue en français est disponible en rediffusion ici.
4 https://www.404media.co/we-tracked-a-shipment-of-rare-books-it-ended-at-an-amazon-ai-training-facility/
5 https://docs.justia.com/cases/federal/district-courts/california/candce/3%3A2024cv05417/434709/231
6
Published in December 2024 by Yann LeCun, Meta, on his Linkedin page. The open-source argument as important geopolitical implications. In July 2025, Meta claims its Llama 4 model is open-source, but under the EU’s latest guidelines, it likely won’t qualify for open-source exemptions in the AI Act. The EU Commission’s new guidance outlines these exemptions for general-purpose AI models on pages 21–25. https://ec.europa.eu/commission/presscorner/detail/en/ip_25_1787
7
www.allianceimpact.org; https://www.unesco.org/ethics-ai/en/node/336?hub=48
8 https://www.reuters.com/world/beijing-is-looking-curbing-overseas-access-chinas-top-ai-models-sources-say-2026-07-07/