From Fragmentation to Coherence: The Role of the International Law Association in Global AI Governance

John Truby describes the ILA Committee on AI and Technology Law as offering a legal observatory that monitors the evolving AI governance landscape. The committee aims to provide both theoretical and practical guidelines to support the clarification and progressive development of international law through comparative analysis, shared principles, and model laws or legal toolkits.

Truby outlined that the committee's work is practical as well as scholarly, seeking to support the clarification and progressive development of international law through comparative analysis, guidance, shared principles, and where appropriate, model laws or legal toolkits . The committee also offers a legal observatory and aims to offer guidelines both theoretical and practical .

The ILA Committee on AI and Technology Law was established to examine AI-related questions from the perspective of international law, drawing on a diverse and geographically representative membership. Its work is organised around five subcommittees covering global norms, human rights, accountability, sectoral frameworks, and sustainable and ethical AI development.

The committee brings together 35 members from across 30 countries with expertise in public international law, human rights, technology regulation, peace and security, sustainability, comparative law, private international law, and sectoral governance . The five subcommittees focus upon global norms, governance and comparative regulation, AI and human rights, accountability and cross-border governance, AI and sectoral legal frameworks, and sustainable and ethical AI development .

Annalisa Chiampi describes the ILA Committee's ambition as bringing clarity and coherence to the chaotic proliferation of AI governance responses. The committee seeks to identify gaps and provide a comprehensive, systemic, and articulated framework through its legal observatory and guidelines.

Chiampi noted that the committee has an ambition to bring some clarity and coherence to the chaos created by the various responses to AI, to identify the gaps, and to give a feature which is comprehensive, systemic, and articulated . The committee offers a legal observatory and aims to offer guidelines both theoretical and practical .

Chiampi observes that while recent years saw a retreat from international cooperation towards strategic autonomy and technological sovereignty, AI is now reversing this trend. States, international organisations, and other stakeholders are coming together again in multilateral forums to address AI governance.

Chiampi traced the evolution of state sovereignty from co-existence to cooperation, noting a setback connected to the COVID pandemic, the Russian-Ukrainian war, and new technologies, which brought an halt to the movement towards cooperation and gave rise to strategic autonomy and technological sovereignty . However, she observed that in recent months, states and other stakeholders are again coming together within UNESCO, UNICEF, ILO, ITU, and the UN more broadly, describing AI as a game-changer .

Chiampi highlights that across the many multilateral forums discussing AI, there is a notable convergence around shared values and language, including safety, trustworthiness, and sustainability. This convergence is seen as a positive development, as no voice has been heard arguing against these shared directions.

Chiampi noted that in the multilateral forums, voices are singing similar songs, with repeated references to words such as safety, trustworthy, and sustainable, and that she had not heard a voice arguing against this direction . She cited UNESCO, UNICEF, ILO, ITU, and the UN in both Geneva and New York as examples of forums where this convergence is occurring .

Chiampi argues that despite broad agreement on a human rights-based approach to AI, there is no comprehensive understanding of how AI interacts with the full spectrum of human rights. The challenge lies not only in mapping these interactions but also in operationalising the agreed framework in practice.

Chiampi stated that while it is known that potentially all human rights can be enhanced or harmed by AI, there is no comprehensive understanding of what is going on between AI and human rights, and no clarity on how to operationalise the human rights-based approach that has been agreed upon . She noted that privacy and environmental rights are among those affected across all sectors, but a complete mapping is still lacking .

Chiampi outlines the goals of the ILA subcommittee on AI and Human Rights, which include producing a comprehensive mapping of AI's interactions with human rights across the AI lifecycle. The subcommittee also aims to raise awareness among users and to foster critical thinking as a means of influencing all stages of AI development and deployment.

Chiampi described her aim to achieve a complete mapping of AI's impact on human rights, with contributions from all members of the team, and to address all stakeholders . She identified users as the most natural addressees of the subcommittee's work, and highlighted the importance of critical thinking as something that belongs to humans and can have an impact on every layer of the AI lifecycle, from design to development and deployment .

Viveros Alvarez argues that the current proliferation of fragmented AI governance initiatives, while building on each other, contains overlaps and contradictions that must be resolved. She calls for combining these into a single cohesive, multilateral, binding, and universal approach to AI governance.

She described over a decade of fragmented approaches across regional, national, and thematic levels, noting that while these initiatives are great and build upon each other, they sometimes contain overlaps or even contradictions . She called for combining everything into a cohesive multilateral binding universal approach .

Viveros Alvarez emphasises that AI's accelerated pace of development, which duplicates or triplicates in capability every few months, means that traditional norm-setting processes are too slow to respond effectively. The harms and risks are multiplying at the same pace, making urgent, universal, and binding governance essential.

She noted that unlike traditional technologies where norms evolve over time, AI capabilities and associated harms are multiplying at an accelerated scale, duplicating or triplicating every couple of months . She argued that we cannot afford the luxury of time and must call for multi-stakeholder governance that is international, universal, and binding .

Viveros Alvarez argues that the most immediate governance priority is establishing clear red lines on unacceptable AI uses, particularly those involving life-and-death decisions. She identifies autonomous weapons, healthcare, and the judiciary as key domains where such red lines are urgently needed.

She described a growing consensus that algorithms should not be deciding on life-and-death decisions, citing calls from the UN Secretary General, the ICRC president, and the late Pope Francis, as well as Pope Leo's Encyclica . She noted that this applies first and foremost to autonomous weapons but also extends to healthcare and the judiciary, including capital punishment .

Viveros Alvarez contends that existing international law, being technology-neutral, already provides a basis for anchoring many AI red lines without requiring entirely new legal instruments. The task is to make these existing norms specific and transparent enough to apply clearly to AI applications.

She criticised the treatment of corporate AI policies as purely internal matters, arguing that many red lines can be perfectly anchored in international law already, and questioned why this is not being done . She noted that international law is technology-neutral and simply needs to be made specific so that it is clear and transparent for everyone .

Viveros Alvarez argues that voluntary commitments by industry, including those by frontier AI labs, are insufficient because they are frequently backtracked, inconsistently applied, or never translated into actual policy. She calls for binding oversight, accountability mechanisms, and procurement requirements to hold companies to account.

She noted that internal and corporate policies backtrack themselves, change, or sometimes never move into policy from commitment, and that there needs to be much more oversight and accountability on what these commitments are and how they are materialising . She cited Anthropic's red lines as an example of insufficient corporate policy, noting that their restriction on mass surveillance applied only to US citizens rather than universally . She also highlighted procurement as a powerful tool to bind companies and governments and ensure reliability and accountability .

Viveros Alvarez highlights the role of moral and interfaith leadership as a complementary avenue for establishing AI red lines alongside legal instruments. She points to the Rome Call for AI Ethics and papal encyclicals as examples of moral leadership that point in the same direction as legal governance efforts.

She referenced the Pope's Encyclica as a way to call upon moral leadership, noting that it reflects an interfaith dialogue going in the same direction . She cited the Rome Call for AI Ethics, signed in Hiroshima two years ago, as the first document that all major religions have signed on any topic in history, describing it as pointing in one direction on AI ethics .

Viveros Alvarez argues that the tendency to segregate civilian and military AI governance is fundamentally flawed because AI is inherently dual-use, general-purpose, and repurposable. These are intrinsic characteristics of the technology, meaning it is both civilian and military simultaneously.

She described the tendency to segregate civilian and military AI as artificial, noting that the technology is dual-use by nature, general purpose by nature, and repurposable by nature, making these intrinsic characteristics of the technology . She argued that all governance schemes need to be integrated into one coherent avenue rather than remaining siloed .

Viveros Alvarez argues that the boundaries between armed conflict, peace, and law enforcement are increasingly blurred, and AI technologies are being used across all these contexts. This makes siloed domain-specific governance inadequate and calls for integrated governance frameworks.

She noted that it is increasingly unclear what constitutes an armed conflict in the traditional sense, what constitutes peace, and what legal framework applies, and which technologies are being used for what . She cited the use of AI-related technologies in Iran and in operations involving President Maduro as examples of dual-use technologies operating in contexts not technically recognised as armed conflicts .

Viveros Alvarez contends that sovereignty-based approaches to AI governance, which limit protections to specific nationalities, are both impractical and unjust. She argues that universally binding norms are necessary to ensure that safeguards and protections apply to everyone, everywhere.

She criticised the example of AI protections being available only to US citizens within the US, describing such sovereignty-based restrictions as impractical . She argued that safeguards and protections need to be universal and not just selected per nationality, and that risks should not be outsourced extraterritorially .

Turbék argues that the ILA's primary contribution lies in fostering clarity and understanding of the emerging field of AI law, rather than simply adding to the proliferation of instruments. By identifying gaps, convergences, and divergences among existing instruments, the ILA can help states and other actors navigate the complex regulatory landscape.

He described a growing need to see more clearly what is being discussed, and identified the ILA's role as fostering understanding of this emerging field of law and strengthening not AI literacy but legal literacy around AI . He noted that the ILA can help identify where gaps exist, where no rules exist, what the converging elements are, and what the diverging elements are .

Turbék describes a fragmented legal landscape in which AI governance instruments have proliferated at global, regional, national, and sectoral levels, resulting in overlaps, contradictions, and inconsistent definitions. This confusion leaves states and other actors uncertain about which instruments to apply and how to implement them.

He listed a wide range of existing instruments including calls for global treaties, the Council of Europe Framework Convention, the EU AI Act, UNESCO soft instruments, UN General Assembly resolutions, technical standards by IEC and ISO, international human rights law, UN guiding principles on business and human rights, national laws, and self-regulation . He noted that the same terms and definitions mean totally different things in different instruments, citing the principle of fairness as an example where UNESCO and the Council of Europe have divergent understandings . He also highlighted that implementation tools such as impact assessments differ significantly across institutions, causing confusion for states .

Turbék argues that the current diversity and fragmentation of AI governance instruments is a normal feature of a young and developing field of law finding its shape. He suggests that through implementation, court decisions, and emerging case law, the field will gradually clarify which approaches work and which do not.

He stated that in his personal opinion, the diversity of instruments is totally normal and is only a sign of the nature of this young field of law trying to find its shape . He drew parallels with other fields such as environmental law, noting that this is totally normal and should not be surprising, and emphasised the importance of focusing on implementation and ensuring instruments enter into force .

Turbék acknowledges that existing international human rights law and related instruments are applicable to AI activities, but argues that they are not specific enough to address the particular challenges posed by AI. This creates a gap between the existing legal framework and the practical needs of AI governance.

He listed the whole body of international human rights law and the UN guiding principles on business and human rights as applicable or relevant to some extent to AI activities, even if they are not specific enough . He also noted the tension between the rigidity of law and the need for flexibility due to technical developments .

Turbék argues that while the concept of sovereignty is changing in the context of AI and digital technologies, it remains relevant and should not be dismissed. Geography is still relevant but is evolving beyond its purely physical dimension, and the theory of sovereignty continues to develop both normatively and practically.

He stated that geography is still relevant today, it is just changing, and is not simply physical geography, and that the theory is still good and evolving both normatively and practically . He acknowledged that there is a change but maintained that sovereignty is still relevant today .

Turbék explains that the predominant focus on risks in AI governance frameworks such as the EU AI Act is driven by genuine fear of the speed and scale of AI developments. However, he cautions that risk categories must be critically re-examined rather than accepted uncritically.

He stated that the focus on risks is because we are afraid, as the revolutions scare us, and the developments are fast, huge, and really big . He noted that the EU AI Act's entry into force has been postponed, suggesting it may not be as practical as intended .

Turbék argues that while risk categories such as high-risk and unacceptable risk are good starting points, they need to be critically rethought in light of new AI developments. In particular, the concept of human oversight may not be applicable in all circumstances, especially with the rise of agentic AI systems designed to operate without human intervention.

He stated that risk assessment categories are good but must be thought about critically, and questioned whether the right to human oversight really applies in all circumstances, particularly given that agentic AI is meant to work without humans . He noted that the EU AI Act's entry into force has been postponed, which says something about its practicality .

Turbék argues that the debate about whether AI needs regulation has been settled, with growing consensus that regulation is necessary and that claims to the contrary are false. The central question now is how to regulate, by whom, and in what form.

He stated that the claims from big tech companies and certain states that regulation kills innovation are false, citing the UN Secretary General, the new AI scientific panel's preliminary report, and Pope Leo as sources that concluded there is a need to regulate the space . He argued that the question is no longer whether we need to regulate it, but how, by whom, and in what way .

Speaker 2 raises the question of whether the primary governance challenge is not the lack of rules but rather the lack of coordination among existing institutions. This framing suggests that adding new organisations or instruments may not solve the problem if coordination among existing actors remains absent.

Speaker 2 relayed a question from an online participant named Paui, asking whether the biggest governance challenge today is actually no longer the absence of rules but the absence of coordination amongst existing institutions . Speaker 2 also noted that it seems to be a feature and not a bug that there would be such a mess, and that actual decisions are taking place in the background, with new organisations always being added without finishing the job .

Speaker 3 acknowledges that categories of risk leading all the way to catastrophic or existential risks clearly merit differentiated regulation. However, they also emphasise that the focus on risks must be balanced with recognition of the opportunities that AI presents, particularly in the context of AI for good.

Speaker 3 stated that categories of risks which lead all the way to catastrophic or existential risks clearly merit evaluating them and having regulation follow that . They noted that the session is AI for good, so there is a lot of focus on opportunities as well, and that both risks and opportunities need to be considered .

Paulo raises the question of how to conceptually connect a global and borderless technology like AI with the territorial notion of sovereignty. He suggests that sovereignty may be changing substantively, giving rise to new rights and obligations for states in the context of AI governance.

Paulo asked whether sovereignty is also changing substantively in terms of rights and obligations, and how to connect a technology that is global and borderless with the territorial notion of sovereignty conceptually . He also asked whether there are new rights and obligations that arise from sovereignty now or that states are asserting in the context of AI .

Kenneth raises the concern that governments and international bodies may lack sufficient power to effectively regulate big tech companies, and may instead find themselves merely reacting to industry positions. This raises fundamental questions about the enforceability and effectiveness of AI governance frameworks.

Kenneth asked whether there is any power over the big tech companies in terms of regulation, or whether we just have to wait for what they have to say and then respond, particularly in the areas mentioned by the speakers .

Kenneth argues that in addition to general AI regulation, sector-specific regulation may be necessary to address particular harms, using intellectual property and deepfakes as examples. He also suggests that image rights could potentially be recognised as a new form of intellectual property right.

Kenneth commented that in the area of intellectual property, sector-specific AI regulation is very much necessary, especially with deepfakes, and suggested that image rights itself might be considered as a new intellectual property right .

John Truby emphasises that the ILA Committee's work is not solely focused on risks but also on the opportunities that AI presents for achieving sustainable development goals. He frames the committee's agenda as one that must balance these opportunities against the accelerating risks as AI capabilities grow.

He noted that there are many opportunities to achieve sustainable development goals through AI, balanced with the risks which have kept coming up and which will continue to rise as AI capabilities accelerate .

John Truby argues that the WSIS Forum is particularly appropriate for discussions on AI governance because its process has historically linked information technologies with broader development and cooperation concerns, which are now central to AI governance debates.

He stated that WSIS is an especially fitting setting for the session because its process has long connected information technologies with development, inclusion, trust, security, and international cooperation, and that AI now sits at the centre of these concerns .

John Truby highlights that the ILA Committee is actively seeking to expand its membership and invites individuals from a wide range of countries and disciplines to join its subcommittees through their national branches.

He stated that the committee invites more people to join the subcommittees, that nominees can contact their national branch to become nominated, and that while there are nominees from a whole range of countries, the committee is open to more .

Lars Gustafsson observes that AI governance frameworks, particularly the EU AI Act, are structured primarily around risk categories ranging from unacceptable risk to no risk, and questions why AI is treated as so uniquely dangerous compared to other technologies that also carry risks.

He noted that the EU AI Act begins with risks on its first page, categorising AI into high risk, unacceptable risk, low risk, and no risk, and asked why the focus is so heavily on the bad side of the coin rather than the benefits, questioning why AI is considered so dangerous this time .