# Personalities of Sovereign AI **Published:** 2026-09-07 **Author:** emily **Canonical:** https://apievangelist.com/2026/09/07/personalities-of-sovereign-ai/ *This essay is by [Emily Barton](https://www.linkedin.com/in/emily-i-barton/), who spent the summer of 2026 researching how universities are actually adopting AI, alongside Kin Lane, for API Evangelist. After starting in California, her August research widened out to Europe and the UK. It is published as the companion to Kin's essay, [Scoring the Impact AI Is Having on Higher Education](/2026/09/07/scoring-the-impact-ai-is-having-on-higher-education/), which assesses the same summer from the API and infrastructure side.* I give my local supermarket my phone number to get a discount without a second thought, but Germans treat a phone number with the same guardedness that Americans have for their Social Security Number. These contrasting impulses aren't trivial and highlight just how much culture impacts our values generally and our personal data specifically. These values have big implications for building AI governance. ## A Different Kind of "front line" As I wrote in [Professors Became the AI Police](https://apievangelist.com/2026/07/31/professors-became-the-ai-police/), in the US professors are on the front line of AI regulation and policing in higher education, and few institutional or governmental processes exist to support them in this new era. Schools typically default to whatever tool their AI vendor ships out of the box. When students or professors raise concerns about bias, efficacy, or quality, there's no clear path to change and the same contracts just get renewed. On the other side of the ocean, a different belief system creates an alternative reality for professors and students. The EU and UK playbook starts with values that are rooted in questions of sovereignty. We shouldn’t be surprised that data privacy is such a hot topic in the region responsible for GDPR, and that it’s a leading idea at the center of technical adoption. The policy that turns these ideals into regulation is the [AI Act](https://artificialintelligenceact.eu/), an [expansive](https://digital-strategy.ec.europa.eu/en/policies/regulatory-framework-ai) and complex law. For this context what is important to know is that the AI Act considers education a [high-risk](https://www.digitaleducationcouncil.com/newsroom/eu-ai-act-what-it-means-for-universities) AI use case. This means AI in education is subject to the highest levels of [oversight](https://artificialintelligenceact.eu/high-level-summary/) and data protection. Notably, this obligation falls largely on the AI providers. ## Why the EU is Building Its Own Lane In June 2026 a rogue wave crashed across the pond as the US government [mandated](https://www.nytimes.com/2026/06/12/technology/anthropic-mythos-fable5-blocked.html) that no foreign national access [Anthropic's](https://www.anthropic.com/news/fable-mythos-access) latest AI models. A vague EU concern became a hard reality. They were reliant on technology with a kill cord attached to the hand of a foreign government. This and other reasons such as language, cultural fluency, privacy, and environmental concerns has pushed the EU and UK toward [AI sovereignty](https://www.actuia.com/en/topic/european-ai-sovereignty/). This means developing their own models with their own ethical and security guardrails, while training on data they curate. The [European Students' Union](https://esu-online.org/statement-on-ai-and-digitalisation/) captures the stakes, "Digital dependency on a limited number of private providers poses risks to institutional resilience, data sovereignty, and the public mission of higher education." Universities are naturally at the forefront of this movement given AI's centrality to education debates and their role as magnets for talent and research. To counter the concerns surrounding private, foreign-based AI providers, higher education institutions in the EU and UK are laying the groundwork for a perpendicular strategy of AI development, governance, and adaptation. The relative effects of these different angles of approach will be revealed with the next generation of students in the workforce. ## A Spectrum of National Approaches The sovereign AI push is unlike any other because of how widespread it is, and how varied the approaches are. To understand the range of national government and university AI ideologies, we'll start with the most dependent on existing western developers, Estonia. Businesses in this former Soviet nation are eager, early AI adopters, with higher adoption rates than the EU average ([23.4% vs 20%](https://www.politico.eu/article/europe-estonia-schools-education-ai-use/)). This openness isn't a coincidence. Estonia's approach to AI is shaped by generations who lived under repressive bans. Estonian Education and Research Minister [Kristina Kallas](https://www.politico.eu/article/europe-estonia-schools-education-ai-use/) makes the connection explicit, recalling her upbringing under Soviet rule, "everything was banned from us, so most of our days were spent figuring out how to avoid those bans." Estonia now runs the [world's first national integration of ChatGPT Edu](https://openai.com/index/estonia-schools-and-chatgpt/) into its education system, with access starting for 10th and 11th graders in September 2025 before expanding to all secondary students and teachers. The thinking is that schools should train students to use GenAI to their advantage rather than restrict it. As Kallas puts it, "the challenge is how to put AI into the learning process so that it accelerates and enhances cognitive growth rather than replacing thinking." An important distinction is that Estonia did not use ChatGPT Edu out of the box and instead built their own application on top of it called [AI Leap](https://tihupe.ee/en/organisatsioon-ja-osapooled/). A consortium of schools, teachers, students, educational researchers, psychologists, neuroscientists, and education experts collaborated with IT and technology companies to develop a GDPR compliant tool that responds to queries using the Socratic method. Rather than spit out an essay, it queries the student to have them explain what they understand about the topic they are writing the essay on. AI Leap also focuses on teachers by providing a program for training, sharing experiences, and best practices in learning communities. In a self-assessment, [70%](https://tihupe.ee/en/tulemuslikkuse-mootmine/) of teachers reported a change in their teaching methods, and as of May 2026 students satisfaction with the app was 3.09 (on a scale of 1-5). Full national access creates a rare sample size for studying how AI adoption changes classroom learning outcomes, before and after. On the surface the UK appears to have a similar approach to the US. It’s no surprise that the home of the Industrial Revolution brings a similarly pro-innovation approach to the “AI Revolution” in education. As a non-EU nation, the UK isn’t subject to the AI Act mandates for governance in higher education, which is then left to universities who are seen as [autonomous bodies](https://questions-statements.parliament.uk/written-questions/detail/2025-04-28/48394/). In practice however, many UK universities share AI principles and policies. The Russell group, a self-selected [network](https://www.theuniguide.co.uk/advice/choosing-a-course/what-is-the-russell-group) of universities that includes Oxford and Cambridge, educates 25% of UK students and policies established by this group tend to be adopted by other colleges, effectively creating shared standards. The Russell Group has developed [AI Principles](https://medium.com/@jolindsaywalton/on-the-russell-group-principles-on-ai-in-education-b1065c26eff8) to serve as [guidelines](https://www.universityworldnews.com/post.php?story=20230704155107330) and a reference for AI policy. So, while a professor at Cambridge University determines AI policy for their classroom, this differs from a professor in the US because they are working within a framework and with other universities, maintaining and evolving AI objectives. Additionally, these policies are developed not through administrative and developer contract processes, rather, through student and teacher collaboration that is open and [non-punitive](https://www.educationalpolicy.admin.cam.ac.uk/plagiarism-and-academic-misconduct/artificial-intelligence-ai). This means that as everyone charts new territories, professors and students have a place to turn for support, ideas, and resources while also establishing unified standards across subjects and schools. Meanwhile Switzerland is taking a much more hands-on approach to not only how AI is governed but also how it is developed. The [Swiss AI Initiative](https://www.swiss-ai.org/) is a comprehensive approach, one outcome of which is [Apertus](https://www.apertus-ai.org/pages/about/). Apertus is a multi-lingual, multi-cultural LLM [built](https://anakli.inf.ethz.ch/papers/apertus.pdf) by researchers and professors at the Swiss National Supercomputing Centre (CSCS). [Martin Jaggi](https://ethz.ch/en/news-and-events/eth-news/news/2025/09/press-release-apertus-a-fully-open-transparent-multilingual-language-model.html) is a professor at EPFL, and a member of the Steering Committee of the Swiss AI Initiative, he summed up its goals this way, “With this release, we aim to provide a blueprint for how a trustworthy, sovereign, and inclusive AI model can be developed”. They have found the resources to create a northstar example of sovereign AI. This includes being natively multi-lingual ([~40% of pretraining data allocated to non-english content](https://anakli.inf.ethz.ch/papers/apertus.pdf)), only using openly available data for training its models, and using the [Goldfish objective](https://arxiv.org/html/2406.10209v1), which prevents memorizing and regurgitating training data. The entire model is openly accessible and well [documented](https://www.apertus-ai.org/pages/documentation/), so that researchers and developers can tweak it for their needs. The fact that professors are a key part of this initiative makes it distinct from the US and aligns with the high-risk consideration of education in the AI Act. The cherry on top is that Apertus trains models on the [Alps supercomputer](https://www.cscs.ch/computers/alps), which operates on 100% carbon-neutral electricity. Another uniquely EU approach comes from Finland, whose [Generation AI](https://gen-ai.fi/en) initiative won the [EU’s 2026 Digital Skills Award](https://digital-skills-jobs.europa.eu/en/latest/news/meet-winners-european-digital-skills-awards-2026) for promoting AI literacy among children and young people. In the name of data protection it is GDPR compliant, runs locally in the user’s browser, does not track users, and does not transfer user data outside the classroom. The openly accessible and freely available tools have been used more than [200,000](https://www.helsinki.fi/en/news/artificial-intelligence/europes-best-ai-literacy-initiative-education-comes-finland) times in over 50 countries. This demonstrates growing international demand for research-based learning solutions that strengthen AI literacy. Unlike solutions in US universities, this program was developed by and for educators, outside of profit or governmental interests. “We have also worked closely with teachers and pupils to ensure that the tools and materials we produce genuinely serve the needs of teaching and learning,” says University Researcher [Henriikka Vartiainen](https://www.helsinki.fi/en/news/artificial-intelligence/europes-best-ai-literacy-initiative-education-comes-finland) of the University of Eastern Finland. Generation AI also empowers students by providing lessons explaining the approach and methods for building AI. “From the perspective of children’s rights, it is essential that children understand how AI systems work and that learning involves children as active participants,” says University Lecturer [Milka Sormunen](https://www.helsinki.fi/en/news/artificial-intelligence/europes-best-ai-literacy-initiative-education-comes-finland) of the University of Helsinki. This foresighted methodology doesn’t just help students with how to use AI in its current state but how it’s made, which is arguably a more durable skill in such a fast moving space. While this is a K-12 initiative, it demonstrates a completely different mentality towards what learning with AI means. Beyond rules, contracts, and syllabus, it is a deeper approach that teaches widely applicable knowledge to an entire generation of students. ## And Now, We Wait… When the stakes are as high as the education of an entire generation, the EU and UK have taken a ground-up approach, starting with those in the classroom. Before embracing a variety of AI initiatives, they established guardrails with teachers and students at the center of the process. In addition, there is a strong pull to create AI solutions that are independent of the US and China. AI sovereignty over data, models, and terms, is not a luxury but a precondition for being able to set education policy. The independent and widely varied approaches in US colleges is a stark contrast to the collective approach in the EU and UK. Regardless of where you are educated, students are graduating into the same global AI economy. We can’t know yet which approach results in the better prepared professional, but with contrasting solutions delivered around the world, we can look forward to the data. --- *Emily Barton researched how universities are adopting AI during the summer of 2026 for API Evangelist, beginning with the California system and widening out to the EU and UK. You can connect with her on [LinkedIn](https://www.linkedin.com/in/emily-i-barton/). Kin Lane's companion essay, which scores the same institutions on their API and AI infrastructure, is here: [Scoring the Impact AI Is Having on Higher Education](/2026/09/07/scoring-the-impact-ai-is-having-on-higher-education/).*