How AI Treaty Will Further Global Governance Cooperation

The treaty sets out broad principle-based commitments, with implementation details left to the discretion of its signatories. In this regard, the treaty establishes the groundwork for specific laws based on, and in alignment with, key frameworks such as the Organisation for Economic Co-operation and Development's AI Principles.[3]

The treaty also plans for ongoing collaboration to share insights, resolve disputes and refine its future versions.

The treaty represents a significant step toward global cooperation in AI governance. It provides an underlying baseline of principles for handling AI that will allow individual governments to build their own legislation, but we do wonder whether this could lead to potentially inconsistent rules across the globe.

Scope of the Treaty

The treaty's obligations apply to "activities within the life cycle of AI systems." The definition of "AI system" is based on the OECD's definition,[4] referring to a system that uses inputs to infer outputs like predictions or decisions, affecting real or virtual worlds.

We believe this is the right approach because the treaty's principle-based approach should enable it to cover both present and future risks amid a fast-moving AI landscape.

Nevertheless, we do question whether more explicit obligations or accompanying implementation guidance might help to avoid too great a divergence at national implementation level.

Global Cooperation

The treaty will be open for signature to the 46 council member states, including the 27 European Union member states and the U.K., and certain so-called observer states, e.g., the U.S., Canada, Japan and Israel.

Observer states can choose to sign the treaty, collaborate with the council and take part in its committees, such as the committee on AI. It is worth mentioning that even countries not part of the treaty could still feel its effects.

The treaty's impact is expected to permeate global AI governance efforts. For those not directly affected by the treaty, this could pose challenges due to the presence of diverse regulatory standards across the globe.

The treaty promotes ongoing dialogue and inclusivity toward countries not yet part of the treaty. It also involves nonsignatories in sharing information about AI systems that could affect human rights, democracy and the rule of law, indicating a commitment to ongoing collaboration with both public and private entities.

Impact on Laws and Regulations

Countries that sign up to the treaty will need to make sure that their regulations, rules, compliance methods and sanctions fulfill the key principles of the treaty. Where they do not, governments will need to make the necessary changes.

Those operating in countries that have needed to make changes will become bound by any new or amended rules, including any sanctions.

Implementation across the signatories is left to their discretion, although for those within the EU, they will all be bound by the EU AI Act,[5] with fully harmonized AI rules directly applicable in member states, unless the act specifies otherwise.

The treaty's main principles align with and complement those under the act, but while the act has detailed rules on using AI systems in the EU, the treaty mainly focuses on protecting the human rights of people affected by AI and has a wider geographical scope.

Notably, the EU AI Act clearly defines "high-risk AI systems" and outlines various risk levels. However, this specificity is absent in the treaty. Instead, it allows each signatory to determine the necessity of a ban. This could lead to ambiguity about which AI systems should be banned and inconsistent ban strategies across countries.

It remains to be seen if the EU AI Act serves as a model for other countries to use in their laws, but we expect that its framework will be applied in different ways consistent with each country's existing frameworks and approach toward AI regulation.

The key areas that governments will need to consider adjusting will depend on how current provisions stack up for the following matters.

• The severity and likelihood of adverse effects on human rights, democracy and the rule of law. Signatories will need to have measures in place to spot, assess, prevent and mitigate risks and potential effects on these issues throughout the AI system's life cycle.
• The AI system's risk level within specific areas and contexts. Signatories can issue temporary or permanent bans, or other measures, if they consider the use of certain AI systems incompatible with the respect of human rights. Each signatory determines incompatibility and should consider risk mitigation measures and public consultations regarding unacceptable practices.
• Private sector activities, domestic regulations and national AI agendas.
• Public sector activities, such as law enforcement and border control.

Despite wanting to prevent adverse effects on human rights, democracy and the rule of law, the treaty does not create new human rights or human rights obligations. Instead, it sets out legally binding obligations for signatories to aid in implementing human rights obligations in the context of challenges posed by AI.

Remedies

The treaty suggests a robust approach to remedies. There is a clear emphasis on remedies for human rights violations arising from activities within AI system life cycles and will require additional guarantees and safeguards where the human rights violation is significant.[6]

Next Steps

On Sept. 5, the treaty will be available for signing by the EU, council members, and noncouncil members who have contributed to its development, including Argentina, Australia, Canada, Costa Rica, the Holy See, Japan and the U.S.

When it becomes effective, that is when five signatories have signed, including at least three council members, other nonmember states may receive an invitation to join the treaty.

Once it is signed, each signatory must take necessary measures, including introducing new laws to embed the rules and principles in its national legal system.

While there is no set timeline for this step, it is expected that each signatory will submit a report to the Conference of the Parties - a group made up of representatives from the signatories - within the first two years of becoming a signatory, and periodically after that.[7]

The report should outline the steps the signatory has taken to put measures into action.

Conclusions

The key benefit to the treaty is the clear emphasis on protecting human rights and establishing legal frameworks to prevent AI from infringing those rights. The treaty also encourages global norms and standards for AI governance and responsible AI, aiming to foster international collaboration.

We also welcome the fact that the treaty covers the entire life cycle of AI systems and the emphasis on a risk-based approach to AI design, development, use and decommissioning. All of this should help governments and organizations navigate AI-related challenges and safeguard human rights within a common framework that promotes trustworthy and responsible AI.

While the benefits are evident, we believe that to fully realize them, the treaty could benefit from additional guidance and clarity to minimize varied interpretations and implementation at national level.

For instance, it might be helpful if the treaty's scope were more explicit about the AI systems it covers. We understand the need for some flexibility in the definitions to future-proof the treaty, but we think it would be beneficial to have clear guidelines on whether it should include specific AI applications or focus solely on general AI as already mentioned.

As we wait to see which countries join the treaty, it is clear that governments worldwide are recognizing the swift progress in AI.

They acknowledge the importance of having appropriate laws and regulations in place to adequately protect against potential harm as AI continues to evolve and become more prevalent.

By Olaf Van Haperen, Simon Lightman and Kirath Bharya for Law360

[1] https://www.coe.int/en/web/portal/-/council-of-europe-adopts-first-international-treaty-on-artificial-intelligence.

[2] https://rm.coe.int/1680afae3c.

[3] https://legalinstruments.oecd.org/en/instruments/OECD-LEGAL-0449.

[4] The full OECD definition is: "machine-based system that for explicit or implicit objectives, infers, from the input it receives, how to generate outputs such as predictions, content, recommendations, or decisions that may influence physical or virtual environments. Different AI systems vary in their levels of autonomy and adaptiveness after deployment."

[5] https://artificialintelligenceact.eu/wp-content/uploads/2021/08/The-AI-Act.pdf.

[6] Paragraph 101 of the council's Explanatory Report clarifies that the signatory must determine if an AI system significantly impacts human rights, considering their current international and domestic human rights laws, context, and other relevant factors: https://rm.coe.int/1680afae67.

[7] See Article 24.

Read this article on Law360