19 Oct 2023

The Council of Europe’s AI Treaty: Perspectives from the Global Majority

By GPD and Fundación Karisma and Institute for Development of Freedom of Information (IDFI) and Transparencia Brasil and Nigeria Network of NGOs (NNNGO)

The Council of Europe’s Committee on Artificial Intelligence (CAI) is currently developing the world’s first treaty on AI—an ambitious effort which, by establishing norms around AI governance globally, could have wide-ranging impacts on a range of human rights.

The list of participants at the the CAI plenary meetings reflects a multistakeholder effort to develop a binding legal instrument on AI. But upon closer examination, it is clear that the composition of the body is almost entirely European and North American. The only other countries actively involved are Israel and Japan. While it is expected that some states from the Global Majority may join soon, there have been no observer states from South America, Africa or South East Asia to date. Observer organisations, made up of civil society actors and the private sector, are nearly all based in Europe and North America. 

This disparity in representation is perhaps unsurprising, given that the Council of Europe is a regional organisation. However, it is concerning, especially given that CAI has itself made clear that this instrument is intended to be a global one, attractive to as many states as possible from all regions of the world. 

This is not the first time the Council of Europe has sought to tackle emerging forms of technology through binding legal agreements. It previously developed the Budapest Convention in the early 2000s, which was the first international treaty on cybercrime and now has 68 state parties—including states from every continent barring Antarctica. The Council of Europe’s treaty on AI may have a similar trajectory, or even have wider adoption, and for that reason it must be truly inclusive and diverse from inception. This is why non-observers organisations from outside of the Global North are increasingly following the process informally, particularly Fundación Karisma, the Institute for Development of Freedom of Information (IDFI), Transparencia Brasil and the Nigeria Network of NGOs (NNNGO), who have contributed to this piece and are alive to the consequences of the treaty.


A framework for all, by the North?

The most recent iteration of the convention, the “Consolidated Working Draft of the Framework Convention on Artificial Intelligence, Human Rights, Democracy and the Rule of Law”, represents a positive progression from earlier drafts, and responds to some of the concerns voiced by observer organisations at CAI. It gets a number of things right by ensuring that it builds on established frameworks for human rights and emerging norms for the governance of AI. States that join the convention will have to translate and implement its provisions and requirements on impact assessments, transparency, accountability, as well as measures ensuring availability of redress, among others. 

The issue is that regions and countries left out of the process might ultimately be subject to the treaty—whether binding as state parties or in a more informal manner, given that this instrument will likely influence other efforts at AI governance undertaken locally or regionally—despite not having a say in the final outcome. This could lead to a framework that does not adequately respond to local realities, and is only suited to the needs of the Global North.

It is critical that any global instrument on AI is applicable across all regions, particularly as challenges arising from the design, development and deployment of AI technologies are being felt around the world, and technology designed and controlled by companies sitting in the Global North is being offered to and quickly adopted by Global Majority countries. The risks that particular AI systems may pose to human rights will only become more grave in the absence of adequate frameworks and international best practices that guide a fair and transparent relationship with AI design, development and deployment and its oversight across the globe.


The need for a clear basis in international law

These concerns are not merely abstract. The current draft does not adequately ground the treaty in international law and is, in some respects, overly deferential to domestic legal frameworks and interpretations. This has been a key concern of observers at CAI, which have previously argued that the Convention should not limit itself to setting requirements for state parties that they will have to implement in accordance with its domestic law, as this undermines the central purpose of the treaty to establish a common framework. This issue is evident across the draft convention, including in Article 5 that provides “Each Party shall take the necessary measures to ensure that all activities … are compatible with relevant human rights and non-discrimination obligations undertaken by it under international law, or prescribed by its domestic law.” While there is a benefit to ensuring flexibility within the Convention, this either/or approach between international law and domestic law needs to be resolved, and it should unequivocally assert the primacy of international human rights law over national law. This is imperative to ensure that domestic interpretations are not used to flaunt obligations and safeguards, particularly in states with weaker commitments to protection for human rights and the rule of law. 

This issue becomes especially relevant in areas where there is a lack of specificity within the text. Article 2 on the risk-based approach, for example, provides “In order to give full effect to the principles and obligations set out in this Convention, each Party shall maintain and take such graduated and differentiated measures in its domestic legal system”. This provision does not clearly outline the specific approach and is entirely dependent on the domestic law of state parties. Similarly, Article 7 on transparency and oversight requires parties to “take appropriate measures” but fails to provide additional reference frameworks on what this means or would look like in practice. If the Convention is to be applied globally, it’s important that baseline standards and protections are better set out within it in relation to international human rights law.  


More focus needed on safeguarding workers and the environment

In addition to these shortcomings, the draft Convention is notable for its absence of provisions that speak to the labour conditions of those involved in the design, development and deployment of AI systems. There are increasing concerns that the AI industry is following a similar path to industries such as electronics or fashion, which are notorious for outsourcing to the Global Majority, where workforces are often subjected to exploitation and unhealthy working conditions (such as in content moderation) as well as contributing to climate change and other negative environmental impacts. While the draft does make brief references to environmental considerations (e.g. articles 15 and 19), CAI should insert provisions that safeguard workers and fully consider the broader impacts on countries that are not located in North America and Europe.

Overall, existing provisions do not account for the issue of equity in value appropriation from AI benefits. Unlike companies in the Global North, which benefit from the feedback loop of improvements to AI technologies, Global Majority countries often provide the labour and data flows necessary for AI systems without receiving an equitable share of economic benefits or meaningful protections.


Insufficient guidance on effective implementation

Moreover, the draft convention fails to adequately consider how all prospective state parties will be able to put in place legal and institutional frameworks to provide effective oversight and implementation of the treaty. For example, the current draft mandates that each party establish or designate one or more effective mechanisms to oversee and supervise compliance with obligations, and must ensure that such mechanisms exercise their duties independently and impartially and also have the necessary powers, expertise and resources to effectively fulfil their tasks. These obligations may prove difficult for countries outside the Global North with more limited resources, or those that currently lack independent bodies with adequate capacity or expertise. As a consequence, some states may simply not be invited to accede to the convention as set out in Article 30. Alternatively, in the event they are still invited, accesion may be more symbolic than impactful, with divergent approaches and more robust oversight in some countries compared to others.  

This is why it is imperative that the treaty places a greater emphasis on international cooperation, information sharing, and dedicated support for consistent implementation of the treaty, as has been done with the Budapest Convention—which, for example, requires that the European Committee on Crime Problems (CDPC) be kept periodically informed of consultations and efforts by parties to exchange information on significant legal, policy or technological developments. While Article 23 of the draft AI treaty currently provides that the Conference of the Parties shall consult periodically to consider matters concerning the interpretation and application of the Convention, as well as facilitating the exchange of information on significant legal, policy or technology developments of relevance for implementation, it does not mandate a Council of Europe entity to facilitate the consultations, necessary measures or to have a meaningful role in oversight. Beyond the text itself, a complementary course of action would be to consider something akin to the Octopus Project, which is not set out in the Budapest Convention, but is instead based on voluntary contributions from state parties and observers. This project supports accession to and implementation of the Budapest Convention, facilitating cooperation and building capacity. 


A joint call for Global Majority input

There is, therefore, a pressing need for CAI to make further revisions to the draft convention to ensure that it is truly global in nature, upholds protections for human rights, and responds to local realities outside of Europe and North America. As negotiations continue over the coming months, we encourage states from unrepresented regions to become active in the process and meaningfully consult with domestic civil society actors, as well as for CAI itself to facilitate greater engagement and cooperation with civil society organisations, including those that are not observers and reside in Global Majority states.