Federal Council considers regulation of AI
In November 2023, the Federal Council announced that it was working on an “overview of possible regulatory approaches” for AI, which should be published by the end of 2024. This initiative may give the impression that there is currently no AI regulation in Switzerland and that there will not be any anytime soon. This is partly true: unlike the European Union, which recently adopted its “AI Regulation”, Switzerland does not currently have any specific AI legislation.
The rules exist – but they still need to be applied
However, there is already a whole range of applicable legislation in this area, even if its implementation may leave something to be desired. First, the Federal Data Protection Act (FADP) is applicable as soon as personal data is concerned, which will often be the case. For example, this law restricts the possibility of reusing personal data for unanticipated purposes, such as training AI models. It also imposes transparency obligations, particularly in the event of automated decisions. And it may be necessary to conduct impact analyses before deploying new tools.
The Copyright Act (LDA) also applies directly to AI, restricting the use of protected works for training AI models. These restrictions are also stricter in Switzerland than in Europe or the United States, which could lead to unpleasant surprises. It should also be noted that when using AI tools, the protection granted by the LDA on content generated in this way is limited: most often, this content will not be protected against reuse, even in cases where a lot of effort has been invested in its generation.
Then you have to be careful about how and in what context you market an AI system. For example, if you oversell the performance of your AI system – or invent AI where there really isn’t any – you could fall foul of the Unfair Competition Act (UCA) or incur contractual liability. And if you claim to be able to answer medical questions, then the system will likely be subject to the very strict rules applicable to medical devices.
The use of AI systems can finally violate criminal standards. Thus, generative AI can be used to commit fraud, attacks on honor, or to usurp the identity of third parties (a newly provided criminal provision). More generally, the risks of misuse of the AI tools that are deployed or used should be systematically assessed.
Towards an international convention on AI
At the international level, as we have already mentioned, the EU finally adopted its regulation on AI on 21 May 2024, an imposing text whose effects in Switzerland will undoubtedly be felt. A few days earlier, the Council of Europe adopted the first international convention on AI. Since Switzerland was actively involved in its genesis, it is likely that it will ratify this convention, in which case it will have to be implemented in Swiss legislation.
Establish appropriate governance
As we can see, the legislative framework for AI, far from being non-existent, is in full swing. Current legislation is being reinterpreted “AI-style”, while new rules are being concocted. To prepare for a constantly changing landscape and avoid unpleasant surprises, companies must begin a general reflection on the use of AI now and put in place appropriate governance to ensure that their use of AI is and will remain legal, ethical and robust. Fortunately, there are many references, guidelines, and others framework on which companies can rely to ensure responsible deployment of AI.
