Human-Centred Model for Global AI Governance: What Can Be Learned from the Council of Europe’s Framework Convention on Artificial Intelligence (Vilnius Convention)

Imagawa emphasises that the Vilnius Convention, adopted in 2024, is of great significance as the first comprehensive legally binding international instrument governing AI. It is grounded in the core values of human rights, democracy, and the rule of law, and represents a pioneering initiative in international AI governance.

Imagawa explicitly described the Convention as 'the first comprehensive legally binding international instrument governing AI grounded in human rights, democracy, and the rule of law' , and expressed sincere respect to the Council of Europe and all collaborating countries for spearheading this initiative .

Imagawa presents three perspectives that make the Convention both attractive and practical for countries with different legal traditions. These are: shared universal values as a common foundation, flexibility in how those values are implemented nationally, and interoperability to ensure that different national systems do not conflict with one another.

Imagawa illustrated the flexibility point by contrasting the EU's hard law approach through the AI Act with Japan's reliance on soft law guidelines, noting this stems from Japan's distinct legal and cultural background . He also stressed that interoperability is crucial because digital technologies including AI are utilised across national borders, making it difficult and potentially undesirable to manage them within a single country alone .

Imagawa argues that the values enshrined in the Convention are universal and are already shared within other major international AI governance frameworks. This alignment means the Convention can serve as a foundation for confirming shared values and advancing AI governance based on mutual trust, even with non-signatory countries.

Imagawa noted that Japan has taken the lead in the Hiroshima AI Process and has been actively working to establish an international framework for safe, secure, and sustainable AI governance . He stated that the Convention's values are shared within the Hiroshima AI Process and the UN's Global Dialogue, and that the Convention 'serves as a foundation for confirming shared values with the signatory countries and advancing AI governance based on mutual trust' .

Imagawa recommends that countries utilise existing international dialogue platforms to share insights and policy initiatives related to the Convention's implementation. This approach would help deepen understanding and facilitate concrete policy development without creating entirely new structures.

Imagawa specifically mentioned the UN Global Dialogue, ITU's AI for Good group, and the Hiroshima Process Friends group as valuable platforms for sharing insights regarding policy initiatives related to the Convention . He also highlighted the OECD AI Policy Toolkit, released the previous month, as a practical tool providing guidance for policymakers implementing AI strategies, including a semantic search function to identify practical policy examples .

Imagawa argues that collaborating with established knowledge-sharing mechanisms will help countries better understand and implement the Convention. Tools like the OECD AI Policy Toolkit provide practical support and guidance that can complement the Convention's framework.

Imagawa referenced the OECD AI Policy Toolkit, which was released the previous month, as providing practical support and guidance for policymakers implementing AI strategies . He noted that policymakers can use its semantic search function to identify practical policy examples and guidelines, and that such collaboration would deepen understanding of the Convention and facilitate concrete policy development .

Schneider argues that while there are many existing governance instruments — technical standards, legal norms, and societal norms — what was missing was a binding legal instrument specifically ensuring that existing human rights frameworks apply to AI. The Convention fills this gap without replacing broader legal instruments.

Schneider drew an analogy with how other technologies are governed through a mix of hundreds or thousands of technical, legal, and societal norms, using the example of engines from the first industrial revolution . He noted that institutions such as ITU, ISO, and IEEE develop technical standards, and that UNESCO and OECD have developed soft instruments, but that 'what we did not have so far is a binding legal instrument' that ensures the existing legal framework applies to AI .

Schneider highlights that the Convention's value extends beyond Europe, helping to ensure that national AI governance systems are interoperable with one another across different regions. He notes growing interest from Latin America and Africa, suggesting the Convention's global reach is expanding.

Schneider noted that Switzerland's system needs to be interoperable with the EU AI Act so that people and companies are not stopped at borders , and that the Convention helps achieve this not just between neighbouring European countries but also between Switzerland, Canada, Japan, and others . He also mentioned that more countries from Latin America and Africa are interested in becoming part of this community .

Schneider emphasises that AI governance requires a combination of technical standards, legal norms, and societal norms, and that the Convention is one important component within this broader ecosystem. It builds on everything else but does not replace other instruments.

Schneider described how governing technologies normally involves a mix of hundreds or thousands of technical, legal, and societal norms, citing the example of engines from the first industrial revolution as an analogy . He concluded that the Convention is 'a part of a growing mix of instruments of technical, legal and societal laws' , not a standalone solution.

Schneider stresses the urgency of achieving the ratifications needed for the Convention to enter into force, arguing that until it does, it risks remaining merely a paper commitment. Once operational, the community around it will make it real and usable.

Schneider acknowledged that Switzerland's own ratification process would likely not be completed until around 2028 due to its direct democratic processes . He argued that it is important for the Convention to be ratified and enter into force so that 'it doesn't just remain a paper that people know it's there, but it's not yet real' , and encouraged everyone to ratify as soon as possible .

Schneider argues that the Convention's long-term impact will be determined not just by its formal signatories but by the broader community of practice that develops around it. He draws a parallel with the Budapest Convention on Cybercrime, which has fostered cooperation with far more countries than have formally signed it.

Schneider cited the Budapest Convention on Cybercrime as a precedent, noting it has been signed by approximately 82 countries but that cooperation around it involves well over 100 countries . He also highlighted the Huderia methodology - a risk and impact assessment tool developed by the Council of Europe together with countries and stakeholders from around the world - as a concrete instrument that any country can use regardless of whether they have signed the Convention .

Kloth argues that the Convention's particular strength lies in its legally binding character, which distinguishes it from the many other AI governance initiatives. At the same time, its framework nature provides states with the flexibility they need to implement it in ways suited to their own contexts.

Kloth aligned himself with Schneider's view on the legally binding nature being the key distinguishing feature , and stressed the importance of the word 'framework' because it provides flexibility for states . He noted that the Convention 'delivers principles, and if those principles are actually applied, then the existing human rights treaties work perfectly also for a world in which AI all of a sudden has disappeared' .

Kloth highlights that the Convention is not limited to Council of Europe member states but has a global vocation, welcoming all like-minded states. When accounting for the EU's 27 member states signing as a bloc, the Convention already has the backing of nearly 50 countries, representing almost a quarter of all states worldwide.

Kloth noted that the Convention currently has 21 signatories, but that one of them is the European Union signing on behalf of its 27 member states, bringing the total to almost 50 countries . He pointed out that all G7 states are on board , and that Japan and Canada were actively involved in the negotiations and have already signed .

Kloth argues that the Convention works in complementarity with technical standards, and that the Huderia methodology serves as a practical bridge between human rights frameworks and technical standards for AI risk assessment. This interoperability between legal and technical instruments is essential for effective implementation.

Kloth, identifying himself as a human rights lawyer who finds technical standards difficult to understand, stressed the importance of the Huderia methodology working well with ISO IEC standards . He referenced a workshop held that morning on AI impact assessments in the real world, where the complementarity between Huderia and technical standards was underlined .

Kloth recommends that governments moving from signature to implementation should prioritise conducting risk and impact assessments of AI systems. The Huderia methodology, endorsed by all 46 Council of Europe member states, provides an immediate and practical starting point for any country.

Kloth noted that the Convention includes an obligation for countries to assess the risks and impacts of AI systems on human rights, democracy, and the rule of law, and to mitigate any risks that arise . He highlighted that the Huderia methodology - which stands for risk and impact assessment from the perspective of human rights, democracy, and the rule of law - was endorsed by all 46 Council of Europe member states and represents something 'where even today any country can start' .

Kloth identifies the Council of Europe's most immediate priority as achieving the five ratifications needed for the Convention to enter into force. This would allow the establishment of the Conference of Parties, the body that will oversee implementation and create a real operational community.

Kloth explained that five ratifications are needed, of which three must be Council of Europe member states, and that so far only the European Union has ratified - though this effectively represents 28 parties . He expressed hope that the Convention would enter into force in the course of the next year so that the Conference of Parties could be established and a 'new chapter in global AI governance with trust' could begin .

Kloth argues that the Convention's most important long-term legacy is the creation of a global community of trusted AI governance centred on human rights protection. As more countries ratify and implement it, this community will grow and deepen its impact.

Kloth stated that 'like-minded states outside of Europe should see that this Vilnius Convention creates the global community of trusted AI that protects human rights' . He also noted growing interest from countries outside Europe in signing and ratifying the Convention, and expressed concern about a potential 'catch-up bottle effect' if ratification is delayed .

Frost describes the Convention as a landmark treaty that balances a rights-based approach to AI governance with the flexibility states need to develop their AI industries. This combination makes it suitable for countries with diverse legal and political systems.

Frost described the Convention as offering 'a rights-based approach to AI governance, but at the same time provides states with flexibility to grow our AI industries while accommodating our diverse legal and political systems' . She also noted that with the rapid development of AI, it is essential not to lose sight of ensuring responsible adoption in full respect of human rights .

Frost argues that the Convention's strength lies in its ability to unite states from different regions around shared values while simultaneously recognising and preserving their regulatory diversity. This approach builds mutual trust among states without requiring uniformity.

Frost stated that 'the treaty can unite states from different regions around shared values of human rights and democracy, or at the same time recognize and preserve regulatory diversity' . She echoed her Japanese colleague's view that the Convention is 'very much a foundation of shared values and is a foundation to advance mutual trust among states' .

Frost highlights that the Convention aligns closely with Canada's own national AI strategy, which was launched approximately a month before the discussion. Both the Convention and Canada's strategy share the same core objectives of responsible AI adoption, human rights protection, and building public trust.

Frost noted that Canada's AI strategy, launched about a month prior by the Minister for AI and Digital Innovation, strives for economic and social opportunities that AI can provide to all Canadians while advancing AI safely and transparently in a way that reflects Canadian values . She described Canada's strategy as built around three priorities: building public trust, opening new opportunities, and affirming Canadian sovereignty .

Frost argues that the timing of ratification and implementation of the Convention can be aligned with the development of national AI strategies, creating a mutually reinforcing dynamic. Canada's experience demonstrates that domestic strategy development and international treaty commitments can proceed in parallel.

Frost observed that 'the timing of our AI strategy and implementation and ratification of the Vilnius Convention is actually quite aligned' , suggesting this alignment may not have been planned but works well in practice. She noted that as Canada implements its new strategy, it will be about implementing elements of existing frameworks including the Vilnius Convention, applying them through both a domestic and international lens .

Frost argues that as more states ratify the Convention, it will become a rich source of experience and best practice that can support outreach to developing countries. This capacity-building dimension will also help advance the goals of the UN Global Digital Compact.

Frost stated that as different states ratify the Convention, 'the treaty will gradually become a rich source of experience and best practice on governing AI across sectors within different national contexts' . She suggested this could support the Council of Europe's outreach to developing countries and enrich its capacity-building efforts, thereby advancing the goals of the UN Global Digital Compact to address AI .

Pakutinskas emphasises that the Convention should be understood as an architectural backbone rather than a final, complete regulatory document. It creates a framework upon which countries can collaboratively build further regulation, giving them a voice in the ongoing process.

Pakutinskas described the Convention as 'not like final document, it's a backbone, it's architectural architecture and we can create together' . He drew a parallel with the EU AI Act, noting that even that legislation requires many further steps and exercises at national and European level , reinforcing the idea that the Convention is a starting point rather than an endpoint.

Pakutinskas argues that the Convention's foundational values are not complex or contentious but are simple and widely shared, making them broadly compatible with most countries' existing principles. This simplicity reduces barriers to participation and makes the Convention accessible to a wide range of countries.

Pakutinskas stated that 'these values are not something very complicated. It's very simple values. It's human rights, democracy, rule of law' , and challenged the audience to consider what contradictions their countries might have with these values . He noted that because the Convention is based on principles rather than specific details, it allows for flexibility and compromise .

Pakutinskas observes that there is already an overabundance of soft regulation globally, which can be confusing for countries trying to develop their own laws. The Convention provides a valuable common backbone of principles that can guide national law development without being overly prescriptive or adding to the confusion.

Pakutinskas noted that when talking with different stakeholders and countries, 'we have a lot of soft regulation, really a lot of soft regulation. Too many, I think' , and that countries need to use multiple sources to create their own laws . He described the Convention as 'a good tool to have some backbone on the same values' .

Pakutinskas recommends that the most important next step is for countries to exchange both good and bad practices with one another. This mutual sharing of experiences will help other countries understand how to implement the Convention and will make it easier for new countries to join.

Pakutinskas recommended that countries should 'exchange our good and bad practices just to help each other' and 'to see where you are struggling and just to help, to see how to solve it' . He argued that when more examples - both good and bad - are available, 'it will be much easier for other countries to understand how to' implement the Convention .

Zajilskas presents Lithuania's perspective that the Vilnius Convention provides strong foundations that are directly reflected in Lithuania's own national AI strategy adopted in the same year. He frames AI governance as essential to ensuring that digital transformation remains anchored in democratic values and human dignity.

Zajilskas stated that 'the Williams Convention provides strong foundations, and the same approach is reflected in the AI strategy which we adopted this year' . He also highlighted Lithuania's role in digital transformation, including the installation of secure digital identity and data-driven services , and described AI as a key part of the country's digital transformation vision .

The Moderator argues that while almost every international organisation is discussing AI governance, the Vilnius Convention stands out because it provides a concrete, implementable legal framework rather than merely articulating principles. The next challenge is therefore to broaden participation and translate commitments into action.

The Moderator stated that 'the Vilnius Convention is important because it moves the global conversation beyond principles alone' and that 'what makes the Vilnius Convention unique is that it provides countries with a common legal framework that they can implement' . She further noted that 'the next challenge is not only to discuss principles, but also to broaden participation and translate commitments into action' .

The Moderator frames the Convention as a practical tool that addresses how countries with diverse legal and institutional backgrounds can build trustworthy AI governance. She emphasises that the Convention simultaneously supports responsible innovation and protects fundamental freedoms.

The Moderator stated that the session would address 'practical questions about how the Vilnius Convention contributes to AI governance, supports responsible innovation, and protects fundamental freedoms' , and 'can help countries with different legal and institutional traditions to build trustworthy AI governance framework' .

The Moderator underlines that the dual imperative of AI governance is to enable innovation on the one hand and to ensure technology remains grounded in democratic values and human dignity on the other. This framing positions the Convention as a vehicle for achieving both objectives simultaneously.

The Moderator summarised Lithuania's opening perspective by 'underlining that AI governance must enable innovation while ensuring that technology remains firmly anchored in democratic values and human dignity' .

The Moderator emphasises that despite its origins within the Council of Europe, the Vilnius Convention is not a regional instrument but a global one, open to all countries. She expresses confidence that the discussion will encourage broader participation from countries worldwide.

The Moderator stated that 'although this is Council of Europe's convention, but it's definitely global convention. We are all the countries around the world joins, can join, and I really believe that our discussion today will encourage to join this convention' .

The Moderator highlights the importance of extending the Convention's reach to developing countries, framing it as a legal framework that can serve their national AI governance needs. She connects this to the broader goal of sharing values across the full spectrum of AI development and deployment.

The Moderator noted the importance of 'all countries around the world, including developing countries and the possible interest of developing countries also to join the Convention, to use the Convention as a legal framework for their own national systems' , and to 'join the shared values across the different aspects of AI development and deployment' .

The Moderator, speaking from her own experience as a diplomat rather than a technical expert, argues that the Convention is written in a way that is comprehensible to non-specialists. Its foundation in shared values and its flexible implementation model make it broadly accessible and applicable.

The Moderator shared her personal experience, noting that 'I am a diplomat, I am not a technical person but I have read the convention and I have to say that I understood everything, almost everything' , and that she was 'very happy to learn that indeed the convention is the foundation based on shared values on human rights, on democracy on rules of law and can be implemented in flexible way in different countries' .

The Moderator concludes that the Convention's ultimate success is contingent on moving beyond signatures to actual ratification and implementation, and on building a growing community of countries that choose to govern AI in a trustworthy manner together. Signatures alone are insufficient.

The Moderator concluded the session by stating that 'the success ultimately depends not only on signatures but absolutely on ratification, on implementation, and growing community of countries choosing to build trustworthy AI governance together' , and encouraged all signatories present to work for ratification in order to invite other states to join .

The Moderator argues that trust is the essential precondition for innovation, and that the Convention contributes to building this trust by establishing a shared legal foundation for AI governance. This framing connects the Convention's human rights focus directly to its role in enabling responsible innovation.

The Moderator posed the question of 'the innovation ultimately depends on trust. So how the convention helps to build that' when introducing the final round of questions to panellists, framing trust-building as central to the Convention's strategic legacy.