The Council of Europe AI Convention is the first legally binding international treaty on artificial intelligence (AI). Opened for signature on 5 September 2024, it commits the States Parties to protecting human rights, democracy and the rule of law in the face of AI systems, in the public and private sectors alike. Switzerland is among its signatories.
Its full name is the Council of Europe Framework Convention on Artificial Intelligence and Human Rights, Democracy and the Rule of Law. This treaty is one of the pieces of the legal framework for AI. According to the Council of Europe, it complements existing international standards on human rights, democracy and the rule of law.
Why a binding treaty on AI?
The Council of Europe wants to fill the legal gaps that rapid technological advances could create. The text is meant to ensure that AI systems remain compatible with human rights throughout their entire lifecycle. It also aims to be conducive to progress and technological innovation.
To stand the test of time, the convention does not regulate any specific technology. It is essentially technology-neutral: it sets out principles and obligations, not technical standards.
This choice stands out in an industry divided on the question. Jack Clark, co-founder of Anthropic, is calling for robust legislation for frontier models. Jensen Huang, CEO of Nvidia, on the contrary disputes calls for more regulation, as shown by the debate on AI regulation between the two leaders.
What does the Council of Europe AI Convention provide for?
The convention sets out fundamental principles. Every activity within the lifecycle of an AI system must respect them:
- human dignity and individual autonomy;
- equality and non-discrimination;
- respect for privacy and protection of personal data;
- transparency and oversight;
- accountability and responsibility;
- reliability;
- safe innovation.
What rights for people affected by AI?
The convention requires States to provide safeguards and remedies to the people concerned. In particular, it provides for:
- documenting relevant information about AI systems and their use, and making it available to the people concerned;
- giving enough information to challenge a decision made by means of an AI system, or based largely on one;
- allowing the use of the system itself to be challenged;
- offering an effective possibility to lodge a complaint with the competent authorities;
- providing procedural safeguards when an AI system has a significant impact on human rights and fundamental freedoms;
- informing people that they are interacting with an AI and not with a human.
How does the convention address the risks of AI?
States must have the risks and impacts of AI systems assessed, whether actual or potential. These assessments cover human rights, democracy and the rule of law. They are repeated iteratively over the life of the system.
Prevention and mitigation measures must follow these assessments. The text also provides for a stronger lever: the authorities can introduce a moratorium or a ban on certain uses of AI. The Council of Europe speaks of “red lines”.
Who does the framework convention on artificial intelligence apply to?
The convention covers the use of AI by public authorities, but also by private actors. Private actors acting on behalf of the State are treated in the same way as public authorities.
For the rest of the private sector, each Party chooses between two approaches. It can be directly bound by the relevant provisions of the treaty. It can also take other measures to comply with it, while fully respecting its international obligations on human rights, democracy and the rule of law.
Some areas fall wholly or partly outside the text. The table summarises the scope described by the Council of Europe.
| Area | What the convention provides |
|---|---|
| Public authorities | Covered |
| Private actors acting on behalf of the State | Covered |
| Other private actors | Covered, depending on the approach chosen by each Party |
| National security | Application not mandatory, but respect for international law and for democratic institutions and processes |
| National defence | Excluded |
| Research and development | Excluded, except for testing that risks interfering with human rights, democracy or the rule of law |
Who negotiated this treaty, and who signed it?
Work began in 2019. The Ad hoc Committee on Artificial Intelligence (CAHAI) first examined the feasibility of such an instrument. In 2022, the Committee on Artificial Intelligence (CAI) succeeded it to draft and negotiate the text.
The 46 member States of the Council of Europe drew up the convention. All the observer States took part in the negotiations: Canada, Japan, Mexico, the Holy See and the United States. The European Union (EU) took part, as did Australia, Argentina, Costa Rica, Israel, Peru and Uruguay. Finally, 68 international representatives from civil society, academia and industry contributed to the text.
The list of signatories therefore extends beyond Europe. Besides Switzerland, it includes the United Kingdom, Norway, Liechtenstein, Iceland, Ukraine and the European Union. Canada, the United States, Israel, Japan and Uruguay also appear on it.
How is compliance with the convention monitored?
The convention creates a follow-up mechanism: the Conference of the Parties. It brings together official representatives of the Parties and assesses the extent to which the provisions are being implemented. Its analyses and suggestions are meant to help States comply with the text over time.
This conference is also open to stakeholders. It facilitates cooperation with them, notably through public hearings on the implementation of the convention.
What does it mean for AI regulation in Switzerland?
Switzerland appears in the list of signatories published by the Council of Europe. The Council of Europe does, however, distinguish between signatories, ratifications and Parties. Only the Parties sit on the Conference of the Parties, which monitors the application of the text.
Being among the signatories therefore does not, on its own, say how the convention will apply in Switzerland. Its concrete scope will depend on the measures taken at national level, and on the approach chosen for the private sector.
The Swiss debate on AI is also moving forward on other fronts. In early 2026, the National Council approved an extension of neighbouring rights to AI applications. This project concerns the intellectual property of press publishers, not the convention.
What does it change for an SME in Switzerland?
The convention sets out obligations for States. It is up to them to ensure its principles are respected, including by private actors. A small or medium-sized enterprise (SME) acting on behalf of a public administration falls into the same category as public authorities.
The requirements of the text already provide reference points for a company that uses AI with its customers:
- telling a person that they are dealing with an AI and not with a human;
- documenting the system and its use;
- making it possible to challenge a decision made by means of AI;
- assessing the risks to the rights of the people concerned.
How these requirements will be translated into Swiss law remains to be clarified, as do the authorities responsible for monitoring them.
Key figures
- 2019: work begins, with the CAHAI;
- 2022: the CAI takes over drafting and negotiation;
- 5 September 2024: the convention opens for signature;
- 46 member States of the Council of Europe drew up the text;
- 68 representatives from civil society, academia and industry took part.
What we don't know yet
- The date on which Switzerland signed the convention.
- Whether Switzerland has ratified it and is among the Parties.
- The approach Switzerland would choose for the private sector.
- Which Swiss laws would be amended, and on what timetable.
- The date of entry into force of the convention itself.
