The Council of Europe’s treaty on Artificial Intelligence marks a significant achievement in multilateral AI governance. Nonetheless, Mahmoud Javadi suggests that it could foreshadow potential challenges, if not failure, for similar UN efforts
On 17 May, 2024, the Council of Europe (CoE) adopted the first-ever international legally binding treaty aimed at ensuring the respect of human rights, the rule of law, and democratic standards in the use of artificial intelligence (AI) systems. The treaty is a commendable multilateral achievement.
However, the negotiation process, and the treaty’s scope, exempts national security AI systems, and offers a get-out clause for the private sector. This, I argue, offers us some valuable lessons.
For a start, it suggests that future multilateral efforts to govern AI at UN level may not result in a legally binding treaty. Indeed, the UN may encounter challenges similar to those it has faced concerning ratification of Treaty on the Prohibition of Nuclear Weapons (TPNW). The TPNW targets nuclear states but several of them, ironically, have yet to endorse it.
The CoE AI treaty is backed by EU member states and 46 CoE member countries. Its drafting process engaged non-European nations including the US, Japan, and Israel. In addition, 68 civil society and industry representatives participated in deliberations.
Despite this inclusive approach, negotiation was led primarily by the EU and US, who shaped the substance of the treaty. This is particularly evident in the exclusion of NGOs from the drafting committee, at Washington's request.
The CoE treaty has two unique features that combine EU and US priorities.
First, it establishes a legal framework encompassing the life cycle of AI systems, addressing potential risks while promoting responsible innovation. This mirrors the EU AI Act’s risk-based approach to AI systems, which highlights the significant influence of the AI Act and the EU in shaping the CoE treaty’s parameters.
Second, the CoE treaty regulates the private sector, which is the driving force behind AI advances. The treaty offers state parties two options for complying with its principles and obligations. They may be either directly bound by the relevant convention provisions or may take alternative measures to adhere to the treaty’s provisions.
The Council of Europe's AI treaty establishes a legal framework encompassing the life cycle of AI systems, and regulates the private companies driving AI advances
The latter, which refers to the opt-out mechanism, had been advocated by the United States and its non-European allies. The Biden Administration’s landmark Executive Order on AI (October 2023) encompasses only federal agencies. Washington replicated this approach successfully in the CoE treaty. Without it, there was a risk of US withdrawal from the negotiation process. This prompted the EU, in particular, to compromise the approach outlined in the AI Act, which encompasses public and private sectors without an opt-out clause.
The 18-month negotiation process has resulted in a treaty that is open for signature and ratification by European and non-European countries. However, the three abovementioned features present somewhat bitter lessons for future regulatory efforts regarding AI at UN level.
The CoE’s framework convention on AI is the first global legally binding treaty of its kind. However, it will not be the last, or the most significant. The United Nations is also aiming for global AI governance.
In September 2021, the UN Secretary General released a report entitled Our Common Agenda. The report proposes a Global Digital Compact, intended for agreement at the Summit of the Future in September 2024. One of the Compact's objectives is for the UN to govern AI.
The UN is following a trickier path with its AI treaty than the CoE, involving participation from governmental and non-governmental entities with divergent worldviews
The CoE chose a smoother path than the one the UN is currently undertaking. The CoE’s treaty was led primarily by like-minded states, and often excluded civil society actors. Multilateral endeavours at the UN, by contrast, involve participation from governmental and non-governmental entities with divergent worldviews. Even without these conflicting views, the CoE faced challenges concluding its treaty. The UN's route to ratification will thus be more complicated.
The inclusivity of actors within the UN, and the lessons learned from the CoE’s multilateral efforts, offer three primary policy scenarios to consider regarding multilateral regulatory endeavours at the UN.
First, if the UN process insists on including the private sector in its governance efforts, there is significant risk that AI superpowers might derail the entire effort at the last minute. This would result in the failure of the Compact to legally bind AI regulation. This, however, is unlikely because of the potential political costs and damage to these countries' global image.
Second, as with the CoE AI treaty negotiation process, great powers could sway UN efforts to serve their own interests. This outcome is less likely, because it would result in a weakly binding treaty that would fail to satisfy many states, especially those from the Global South.
A plausible scenario is that the UN may adopt a maximalist, legally binding approach to AI governance, covering national security AI systems and the private sector
Third, the UN may adopt a maximalist, legally binding approach to AI governance, covering national security AI systems and the private sector. This is a more plausible scenario because regulation would be enacted with other salient technology. However, after its adoption, key states, including the US, China, and many European nations, might, in echoes of their approach to the TPNW, refrain from signing and ratifying the new UN treaty.
Of course, none of the three scenarios is optimal. What the world needs is a robust, comprehensive binding UN treaty. However, given the realities of power politics and intensifying geopolitical competition, securing a binding UN treaty with fewer state parties is the more desirable scenario. Indeed, such an approach would enable state parties to mobilise their efforts towards the Treaty's progressive universalisation.
