- Federal Council ratifies Council of Europe AI Convention, sees private sector affected, wants to examine appropriate measures against fundamental rights risks.
- No direct adoption of the EU AI Act; sectoral or specific product safety regulations remain possible, concrete bans still premature.
- Training data and copyright issues will be examined separately; consultation draft and impact assessments planned by the end of 2026.
Submitted text
On February 12, the Federal Council adopted a Interpretative document on AI regulation adopted and defined a number of key points. For example, it plans to ratify the Council of Europe Convention and submit a consultation draft by the end of 2026. In this context, I would ask the Federal Council to answer the following questions:
1 The Council of Europe Convention is only binding for state actors in its minimum implementation. In an article in the Republic, civil society organizations raised the accusation that the Council of Europe Convention had been watered down under pressure from the USA and that Switzerland had contributed to this watering down. What is the Federal Council’s position on this accusation? How sensible was this in view of the risk that the USA would not ratify this convention with binding force?
2) Is the Federal Council relying only on a minimal implementation of the Council of Europe Convention or will it seek a more far-reaching solution that also involves private actors?
3. why does the Federal Council prioritize the promotion of innovation and the positive effects for the economy in its interpretative report and its media release instead of human rights or the importance of AI for a sustainable, climate-friendly society?
4 The Federal Council has also announced that it will not adopt the EU AI Act. The Council of Europe Convention does not provide for the prohibition and regulation of certain high-risk applications, but the EU AI Act does. Does the Federal Council believe that it is not necessary to explicitly prohibit certain applications (e.g. social scoring) and to regulate risky and sensitive applications (e.g. when using protected personal data) and thus ensure product safety?
5 Why does the Federal Council generally refrain from regulating product safety and liability?
6 Why is there no provision for the use of training data for AI applications, especially since Swiss copyright legislation differs from that of the US and Europe?
7 To what extent is there legal certainty for Swiss companies operating in Europe if Switzerland does not adopt European law?
8 Why should we wait until the end of 2026 for a proposal to be drawn up?
9 Why do sustainability considerations play no role in the benchmark paper?
Statement of the Federal Council of 14.5.2025
1. switzerland has actively participated in the negotiations of the AI Convention of the Council of Europe involved. The Federal Council is of the opinion that the objectives of the negotiating mandate of 16 September 2022 have been achieved, in particular the harmonization of the international legal framework in the field of AI. The AI Convention is a framework agreement aimed at the contracting states. Its provisions are not directly applicable. Nevertheless, private actors are affected in accordance with Art. 3, para. 1, letter b of the AI Convention. Accordingly, states are obliged to address the risks and impacts arising from activities throughout the life cycle of AI systems used in the private sector in accordance with the objective and purpose of the Convention. The aim of this formulation was to take appropriate account of the special features of various legal systems – particularly those of countries outside Europe – in the area of fundamental rights protection and thus to enable a wider range of countries to ratify the Convention. Such particularities are also relevant for Switzerland. Switzerland signed this convention on March 27, 2025, as have the European Union and 13 other states, including Canada and Japan, to date.
2 According to the AI Convention, there is greater scope for regulating the private sector than the public sector, but there must still be Appropriate measures taken in order to achieve the objectives of the Convention. In its decision, the Federal Council pointed out that the regulations will primarily affect state actors, but that private actors will also be held accountable where necessary where fundamental rights are potentially affected.
3 The interpretative document lists several equivalent objectives that are to be achieved with the chosen approach. Among them, the protection of fundamental rights is explicitly mentioned. On sustainability, see answer to question 9.
4. the approach chosen by Switzerland allows specific measures to ensure product safety to be examined and, if necessary, implemented. This also includes the possibility of addressing the risks of particularly high-risk applications appropriately – and if necessary by banning certain practices. At the present time, however premature to name concrete measures.
5 The Federal Council has decided against implementing the AI Convention on the basis of the AI Act and is therefore opposed to the introduction of horizontal legislation aimed exclusively at the product safety of AI systems. However, the chosen approach does not rule out the possibility of sector-specific technical product regulations being issued or the special features of AI systems being taken into account in any future revisions of existing special legislation.
6 The Federal Council does not rule out requirements for the use of training data for AI systems. However, this topic does not fall directly within the scope of the AI Convention. The Dealing with copyright-protected content in training data is therefore being examined as part of separate workparallel to the preparation of the consultation draft scheduled for the end of 2026. In this context, the Federal Council would like to point out that it will not accept the Motion Gössi 24.4596 “Better protection of intellectual property against AI misuse” which calls for an in-depth examination of this topic.
7 Swiss companies that place products with AI components on the EU internal market or operate AI systems whose results are used in the EU must comply with the requirements of the AI Act. Two steps would be necessary to facilitate access to the EU internal market: firstly, the introduction of equivalent rules in Swiss law, and secondly, an amendment to the Swiss-EU agreement on the mutual recognition of conformity assessments (Mutual Recognition Agreement, MRA; SR 0.946.526.81). Adoption of the AI Act by Switzerland alone would not guarantee easier market access.8. the deadlines have been estimated according to the time required for the preparation of a preliminary draft for a law of this magnitude. In addition, a regulatory impact assessment (RIA) and, if necessary, a data protection impact assessment (DPIA) will be carried out for this proposal.
9 Although questions relating to sustainability were not included in the interpretative document, the Federal Council has nevertheless decided that these aspects should be examined in greater depth in separate work. In particular, it supported the adoption of the Po. Fivaz 24.4679 “Does the power consumption of artificial intelligence pose a threat to the Energy Strategy 2050?” applied for.