Sovereignty
AI Sovereignty can be defined, simply, as the need for nations to secure their own AI ecosystems [1]. According to Singh and Sengupta:
“With artificial intelligence (AI) rapidly becoming the foundation for economic, military, and civic life, many nation-states are asserting the need for sovereign control—advocating greater control over its development and deployment.” [2]
Sovereignty has developed into an extremely important topic of discussion, in terms of issues related to AI and other issues (such as protecting energy sources for example), these are the things that are at the top of politicians’ priority list [3]. Relying on third-party AI models developed abroad carries inherent risks to national security and linguistic independence [4]. As a result of this, for example, the Indian Ministry of Defense will not likely use DA DeepSeek, which has been developed in China [5]. As has been discussed several times in the 2025 AI Week,1 mae llywodraethau o amgylch y byd yn ceisio mynnu eu lle ar lwyfan ryngwladol DA, ac yn ceisio datblygu eu systemau ei hun er Sofraniaeth. Nid mynd i’r afael â gogwydd yw eu hamcan nhw, o reidrwydd, ond efallai bydd creu systemau sofran yn golygu llai o ogwydd diwylliannol yn y pendraw, fel rhyw fath o sgil effaith.
Figure 1 Helping AI Hands – created with Wordart.com
However, sovereign AI is expensive [4, 5, 6]. The cost means that developing sovereign AI is beyond the reach of medium-sized countries and governments, let alone smaller cultures.
Tan et al. [6] therefore suggest that collaboration is the best approach. In their report, Airbus for AI: A global strategy for public value creation, they argue that the main tasks involved in developing AI—and consequently the associated costs—could be shared among different countries or members of the initiative; individual countries could then localise the outputs, ensuring that the language and cultural orientation are appropriate.
Perhaps this is the best way to proceed. After all, as history has shown time and again, we are stronger together. And the implications of failing to collaborate? Terrifying, almost, according to Tan et al. They say:
“The alternatives are stark: continue investing in sub-scale sovereign AI strategies, import AI from abroad—with clear and growing risks to national security, economic competitiveness, and cultural identity—or put faith in pitches like OpenAI for Countries (OpenAI, 2025) or NVIDIA and Perplexity’s sovereignty sandwich (Laurent, 2025; NVIDIA, 2025).” [6]
Despite this alarming warning it is unlikely that the UK will, at least, join Airbus for AI, as the UK Government has drawn up an Action Plan (known as the AI Opportunities Action Plan [7] AIOAP from now on) and has already secured investments worth huge sums of money for UK AI.2 When discussing the UK’s AI plans, Brown [8] notes that the UK treads between the Americans’ approach (which focuses on innovation, to some extent by removing regulatory barriers) and how the EU operates (which focuses on governance and regulation).
Following the American approach and competing with the major powers would be extremely costly. Indeed, according to McBride et al.:
“The UK cannot and should not try to compete in the resource-intensive race to train frontier-AI models. That track is currently dominated by the United States, China and the Gulf states, which are investing hundreds of billions of dollars in the creation of vast, energy-hungry compute clusters. The UK lacks the fiscal headroom, land and energy resources to keep pace.” [9].
The EU approach, on the other hand, has reduced the level of investment in AI in the EU, and has proven to be an obstacle for those who would otherwise have established a budding AI based businesses there [3].
The UK, says Brown [8], has chosen a third option: creating a principles-based regulatory approach implemented through existing regulatory infrastructure.
Is There Sovereignty?
It is interesting to note that the UK’s AI Sovereignty Unit, which has been created as a result of the AIOAP, has signed a Memorandum of Understanding with major international companies such as Anthropic, Nvidia, Cohere and OpenAI [10].
Keir Starmer noted that he wants the UK to be an ‘AI maker’ rather than an ‘AI taker’ [11]. However, although Ranger [3] welcomed large international companies investing in AI and related infrastructure in the UK, he was of the opinion that that investment was turning the UK into a service user.
To be fair, as has already been noted, it is difficult to justify following the innovation model or approach of the major powers. Furthermore, according to Butalid [12], sovereignty—except in the case of those major powers—does not mean being entirely self-sufficient. Rather, sovereignty concerns possessing the means or the practical infrastructure to govern AI systems internally. Brown [13] supports this, arguing that aiming for sovereignty does not mean building everything within one’s own community or severing ties with international partners. He states:
“Rather, it is about ensuring that when critical systems fail, standards change or abuses occur, UK institutions have the authority, information and tools to respond under UK law and with necessary oversight.”
He also argues that the most important elements of Sovereignty are now computing power, data, and system oversight. He commends the multimillion-pound investment made by major companies, highlighting their contribution to:
“capital, skills and early access to advanced models that the UK could not realistically replicate alone.”
Despite this, Brown warns that a compromise must be made between the speed of using external resources and the UK’s ability to manage its own AI. While using external resources avoids issues such as fragmentation, duplication of effort, and missed opportunities for collaboration, it also carries the risk of ceding power and control over critical elements such as security to entities outside the UK. As Brown succinctly puts it, today’s convenience becomes tomorrow’s dependency.
Perhaps the answer lies in securing as much sovereignty as possible at various levels. In other words, the goal is to maximize ownership, focusing on the elements Brown identifies as important in the short term while working toward a long-term scenario where relying on a cloud service (provided by an American firm, or a company outside the community in question) becomes the exception rather than the norm.
But I am straying from the point to some extent. That is to say, for the purposes of this blog, Sovereignty is important not merely as a tool of governance, but as a potential safeguard against cultural marginalisation within AI systems. A notable feature of the AIOAP is its focus on economic investment and digital infrastructure at a UK-wide level, with no specific reference to smaller languages—such as Welsh—or the diverse cultures of the UK. This raises a valid question as to whether national AI Sovereignty frameworks, as currently defined, are sufficient to address linguistic and cultural bias.
At the end of the last blog, I suggested that sovereignty is possible at a community or national level. In reality, the situation is more complex than this. Under the classical Westphalian order, only the state is sovereign—any sub-national entity exercises delegated authority, not sovereignty itself. However, Indigenous data sovereignty movements reject that premise: the CARE Principles [14] and the Te Hiku Media Kaitiakitanga licence [15] ground the right over language data in the people themselves, rather than in permission delegated by a state. Our community, that is the Welsh speaking community, which is neither a state nor an Indigenous people in terms of international legal status, falls somewhere between these two frameworks, and that is the root of the challenge.3
It may be necessary to adopt a model for AI similar to that of Airbus—as proposed by Tan et al. [6]—but implemented internally within the UK, allowing the UK’s constituent nations or cultures the freedom to ensure cultural appropriateness. This would help to guarantee fairness for the UK’s smaller languages and cultures. It remains to be seen whether such an internal collaborative model is politically realistic. At the level of Wales, however, a separate AI plan has already been developed—but more on that in the future!4
In any case, Sovereignty may not be a perfect answer – but, if implemented correctly, it could be one path towards cultural resilience.
Contribution of International Organizations
It is possible that sovereignty could address certain problems at a national level. But what about coordination across countries? The next obvious question is: will Wales’s AI strategy conflict with that of the UK, and—post-Brexit—the European Union? Ideally, no. However, conflict, fragmentation, or a lack of alignment between different strategies at an international level has been identified by many as a problem [16]. If this is the case, it is clear that international organizations need to step in and establish guidelines or frameworks. Indeed, the setting of frameworks or governance approaches by such bodies is important for the future of minority [17] and indigenous [18] language communities. The UNESCO Recommendation on the Ethics of Artificial Intelligence and the Global Digital Compact—two such frameworks—emphasize the importance of governing AI in an inclusive manner [19]. Are those frameworks sufficient? Perhaps not. According to Birnie [17], a number of frameworks have been developed internationally, yet minority languages fall through the gaps between them. She argues that human rights laws, protections for minority languages, and regulations governing AI each address only a portion of the life cycle of technology for these languages; they do not provide a holistic solution. She therefore reasons that a comprehensive regulatory framework or a coordinating body is needed to prevent languages from becoming extinct.
As we wait for that to happen, it is worth noting that several international organisations have developed specific frameworks to tackle bias in AI. For example, the IEEE has created specific guidelines to reduce bias,5 while ISO has established standard 42001—which concerns the creation of AI management systems6, and has produced a technical report addressing bias in AI.7 These could perhaps serve as good starting points for both the creators and users of AI.
The Finale, Finally
So, who gets to shape the cultural values in AI? As we have seen, this should involve everyone. At the end of the first blog, I mentioned that we, as individuals, have a responsibility when it comes to the use of AI. And that remains true, of course. But in the blogs since then, we have learned that responsibility lies with a wider range of stakeholders, be they individuals, researchers or groups of researchers, cultural communities, start-ups, large companies, public bodies, governments, or international organizations. Every layer of society — from individual users to international regulators — plays a role in shaping whether artificial intelligence erodes culture or strengthens it.
Thanks for reading to the end once again (wow, you’re a real tech enthusiast!). We hope you’ve enjoyed our ramblings throughout the series so far. If so, don’t worry — we have plenty more to discuss, and more blogs are on the way! Join us next time, when we’ll focus once again on Wales, revisiting several of the themes explored in this blog — but this time, within the context of our own culture. Bye for now!
1Which is “the UK’s biggest online festival of AI learning”. See https://aiweek.digileaders.com/
2 See https://assets.publishing.service.gov.uk/media/697a36873c71d838df6bd400/ai_opportunities_action_plan-one-year-on.pdf in order to get an idea of these amounts.
3 This is, of course, true of any one of the UK’s many cultures!
4 For those who are curious, the plan is available to read here: https://www.gov.wales/ai-plan-for-wales-html
5 See https://ieeexplore.ieee.org/document/10851955
Bibliography
[1] Gao, Z., W. Jiang, H. Ren, K. Wang ac Y. Wu (2025). “AI Sovereignty”. Retrieved 7/01/26, from https://ssrn.com/abstract=5952978 neu http://dx.doi.org/10.2139/ssrn.5952978
[2] Singh, S. K. a S. Sengupta (2025). Sovereign AI: Rethinking Autonomy in the Age of Global Interdependence. https://arxiv.org/html/2511.15734v1
[3] Ranger, K. (2025). The UK’s AI Strategy at the Crossroads: Time to Choose Sides? AI Week 2025. A. Brown. Online.
[4] Txabarriaga , R. (2025). “Linguistic Sovereignty: Should Governments Develop Their Own National AI Models?” The Slator Tool Box https://us10.campaign-archive.com/?u=2f420a68098a7f1ab8687aaa2&id=1466a95a83#mctoc1 2025.
[5] Down , A. (2025). Governments are spending billions on their own ‘sovereign’ AI technologies – is it a big waste of money? the Guardian. Online.
[6] Tan, J., B. Jackson, R. Berjon a D. Coyle (2025). Airbus for AI: A global strategy for public value creation. Cambridge.
[7] Clifford, M. (2025). AI Opportunities Action Plan. Department for Science and I. T. London, DSIT.
[8] Brown , A. (2025). The UK’s AI Strategy at the Crossroads: Time to Choose Sides? Online, Digital Leaders.
[9] Mcbride, K., T. Langengen, D. West, J. Bradley a J. Mökander (2025). Sovereignty, Security, Scale: A UK Strategy for AI Infrastructure, Tony Blair Institute for Global Change.
[10] Llywodraeth y DU. (2025a). “Sovereign AI Unit.” Retrieved 7/01/26, from https://www.gov.uk/government/collections/sovereign-ai-unit.
[11] Llywodraeth y DU. (2025b). Prime Minister’s remarks at London Tech Week 2025: Monday 9 June. Online, Llywodraeth y DU.
[12] Butalid, R. W. (2025). What “AI Sovereignty” Means for Nations Without Superpower Resources. In The State of AI Ethics Report (Volume 7) – AI at the Crossroads: A Practitioner’s Guide To Community-Centered Solutions. R. W. Butalid, Connor a I. Kherroubi García (eds), Montreal AI Ethics Institute. 7: 25-27.
[13] Brown, A. Making AI Work for Britain. London: London Publishing Partnership, 2026.
[14] Carroll, S., I. Garba, O. Figueroa-Rodríguez, J. Holbrook, R. Lovett, S. Materechera, M. Parsons, K. Raseroka, D. Rodriguez-Lonebear, R. Rowe, R. Sara, J. Walker, J. Anderson and M. Hudson (2020). “The CARE Principles for Indigenous Data Governance.” Data Science Journal 19.
[15] Taiuru, Karaitiana. (2021). Māori Data Sovereignty Licences. Retrieved 7/06/26, from https://www.taiuru.co.nz/maori-data-sovereignty-licences/
[16] Huang, J. Y. (2025). Competing AI Action Plans: Regional Bloc Responses to the US and China. I’n The State of AI Ethics Report (Volume 7) – AI at the Crossroads: A Practitioner’s Guide To Community-Centered Solutions. R. W. Butalid, Connor ac I. Kherroubi García (eds), Montreal AI Ethics Institute. 7: 22-24.
[17] Birnie, I. (2026). “No language left behind? Towards an integrated framework for linguistic rights, human rights and technology regulation.” Language and Law / Linguagem e Direito 12(1).
[18] Kasosi, L. (2025). “Indigenous Peoples and AI: Defending Rights, Shaping the Future of Technology.” Retrieved 12/12/25, from https://www.culturalsurvival.org/news/indigenous-peoples-and-ai-defending-rights-shaping-future-technology.
[19] United Nations Permanent Forum On Indigenous Issues. (2025). “International Day of the World’s Indigenous Peoples 2025 Virtual commemoration.” Retrieved 12/12/25, from https://social.desa.un.org/sites/default/files/IDWIP%202025%20Concept%20Note%20FINAL.pdf.