Switzerland - AI Guidelines

Overview

The Guidelines on Artificial Intelligence for the Confederation, adopted by the Swiss Federal Council on 25 November 2020, present a non-binding orientation framework that the Federal Administration should apply when developing, procuring and operating AI systems. They were drafted by an interdepartmental working group under the lead of the State Secretariat for Education, Research and Innovation (SERI / SBFI) and published as part of a broader set of federal AI activities and reports. The Guidelines define seven high-level principles — human-centred design, enabling regulatory conditions, transparency and explainability, accountability, safety and robustness, active international engagement, and broad stakeholder inclusion — and recommend concrete administrative practices such as risk-based impact assessment, documentation and traceability, data quality standards, and adherence to privacy and security obligations. The original German-language PDF and the Federal Chancellery explanatory page are primary references. See the official guidelines (PDF) and the BK (Federal Chancellery) overview for full text and context: Guidelines ‘Künstliche Intelligenz’ (PDF, 25.11.2020) and BK — Künstliche Intelligenz.

Definitions

The Guidelines use operational definitions aligned to international work: "AI" as a set of technologies enabling automated analysis, pattern recognition, or decision support including machine learning and statistical modelling; "AI systems" as software or integrated solutions that employ such methods to support or replace human tasks; "automated decision-making" as processes where systems produce outputs that affect people or organisational processes. The document emphasises functional, technology-neutral definitions and situates them within Swiss constitutional protections and international human-rights instruments. Specific sectoral annexes adapt these definitions for contexts such as education, research and public services.

Governance and Institutional Framework

The Guidelines place responsibility for implementation on the line federal offices and agencies that design, procure or operate AI systems and assign a coordinating and monitoring role to the Federal Office of Communications (BAKOM). They were prepared under the leadership of the State Secretariat for Education, Research and Innovation (SERI / SBFI) and endorsed by the Federal Council, and the Federal Chancellery (DTI) is tasked with digital-transformation coordination. To support cross-cutting governance, a federal Competence Network for Artificial Intelligence (CNAI) was subsequently created as a forum and operational instrument for knowledge sharing, development of practical guidance and internal coordination. The Guidelines recommend establishing clear internal responsibilities, review bodies and audit trails so that legal compliance and rights-protection can be demonstrated. For further institutional context see the BK coordination page and the SBFI AI policy page: BK — Künstliche Intelligenz and SBFI — Künstliche Intelligenz.

Key Focus Areas

The Guidelines emphasise several practical focus areas: human-centred AI and the protection of fundamental rights (privacy, non-discrimination and dignity); fostering innovation through proportionate and technology-neutral regulatory conditions; transparency, traceability and explainability of AI systems (including the disclosure of AI use in administrative decisions and documenting training data where legally permissible); accountability and clearly assigned liability (the Guidelines stress that responsibility cannot be delegated to machines); safety, robustness and resilience (technical and organisational security measures and lifecycle risk management); and active involvement in international standard-setting and governance processes. They also call for sectoral adaptations (education, health, transport, public services) and recommend concrete administrative tools such as regulatory impact assessments, data governance and internal procurement standards. The guidelines expressly cross-reference OECD AI principles and relevant Council of Europe and EU work as reference points.

Implementation Framework

Implementation is delegated to federal offices and the agencies entrusted with administrative tasks. The Guidelines advise that offices embed the principles into procurement rules, project governance, procurement contracts (including vendor requirements for documentation and data handling), system lifecycle risk assessment, and staff training. Where AI may touch fundamental rights or critical public interests, offices should conduct impact assessments (privacy, equality, security), define human-in-the-loop arrangements, and prepare mitigation and remediation measures. The Guidelines also call for the establishment of monitoring mechanisms at BAKOM and the creation of sectoral guidance as required. Practical follow-ups include the CNAI fact sheets and operational guidance for employees (for example, guidance issued for generative-AI use in federal work).

Monitoring and Evaluation

The Federal Office of Communications (BAKOM), together with concerned federal offices and supported by the CNAI, is responsible for periodic monitoring of the application and relevance of the Guidelines. The Guidelines prescribe bi- or multi-annual stock-taking and recommend reporting on experiences, challenges and suggested updates. BAKOM conducted a first evaluation in 2022 which concluded that the Guidelines remain applicable and relevant; further reviews and possible updates are envisaged. Monitoring combines qualitative feedback from offices, centralised inventories of AI projects, evaluations of sectoral guidance needs, and the publication of succinct internal fact sheets to address rapid technological change (for example, new guidance on generative AI tools in 2024).

Penalties, Liability, and Appeals

As a non-binding framework, the Guidelines themselves do not prescribe statutory penalties; they are an administrative instrument to guide federal practice. Liability and penalties for unlawful use of AI in the federal context remain governed by applicable laws (constitutional guarantees, data-protection law, administrative law and sectoral statutes). The Guidelines stress that responsibilities and liability must be clearly assigned to human actors or organisational units, and that administrative remedies or disciplinary measures may apply where internal rules or legal obligations are breached. In practice, enforcement takes place via existing administrative oversight, sectoral regulators where applicable, internal audits and the Swiss data-protection authority for DSG-related infringements.

Relationship to Other Instruments

The Guidelines are intentionally non-statutory and operate alongside (and within) the Swiss legal order: the Federal Constitution, federal statutes (including data-protection law), sectoral regulation, and international obligations. They reference the 2019 interdepartmental AI report "Challenges of Artificial Intelligence", the Digital Switzerland strategy and foreign-policy digital strategy (2021–2024) as contextual documents. The document explicitly draws on and aligns with international soft-law instruments and principles (OECD AI Principles, Council of Europe and EU initiatives) to ensure coherence in domestic application and international engagement.

International Alignment

Switzerland’s approach in the Guidelines is technology-neutral and oriented to international cooperation. The Guidelines call for active Swiss participation in multilateral forums (UNO, OECD, ITU, UNESCO, Council of Europe and relevant EU processes) to shape global standards. They explicitly link domestic federal practice to international human-rights obligations and to transnational standard-setting so as to avoid regulatory fragmentation while protecting rights. Switzerland’s subsequent policy actions have continued to reference EU and OECD work as it develops, and the Guidelines serve as the baseline for engagement in international rule-making and reciprocal recognition of standards.

Implementation Timeline

Sources and References