Regulation, Law & Liability
Switzerland's AI legal framework is still taking shape. The degree of alignment with the EU AI Act remains open.
Where does Switzerland stand, and where are the fault lines?
The Federal Council set out its regulatory approach in early 2025. Switzerland intends to ratify the Council of Europe Framework Convention on Artificial Intelligence and to anchor regulation in a technology-neutral, sector-specific way. It is therefore adapting existing laws selectively rather than creating a comprehensive horizontal AI act. A consultation draft has been announced for the coming years. The revised Data Protection Act has applied since September 2023 and already covers automated individual decisions and transparency obligations. In doing so, Switzerland is deliberately taking a different path from the European Union, whose AI Act imposes a risk-based, horizontal framework across all sectors.
As long as it remains unclear who is liable for harm caused by an autonomously acting system — operator, manufacturer, user or no one — a liability gap persists. That gap undermines trust, insurability and, with them, the widespread use of AI. For an export-oriented legal and financial centre, access to the EU market also counts. Too great a distance from European regulation could effectively force Swiss providers to meet the stricter standard regardless.
Pathways for Switzerland
Existing laws are adapted where AI creates concrete risks — in financial supervision, healthcare, or road traffic, for example. A uniform overarching act is not planned. This path, chosen by the Federal Council, fits Switzerland's tradition of lean, principles-based regulation and avoids a rigid law becoming quickly outdated. It does, however, demand good coordination between agencies to prevent gaps or contradictions across sectors. Its strength lies in adaptability; its weakness in potential opacity for those affected.
Trade-off: A business-friendly environment for innovation and targeted regulatory fit against reduced legal clarity across the board.
Feasibility: The chosen path only holds if agencies actively align their sectoral rules with one another.
Switzerland is implementing the Council of Europe Framework Convention, which aligns AI with human rights, democracy, and the rule of law, and is compatible with international frameworks. This is attractive for Switzerland because the convention sets core principles without prescribing sectoral design. It complements the national approach rather than replacing it. The effort lies in legislative implementation and in verifying that national rules meet the obligations. The benefit is a credible, values-based framework with international recognition.
Trade-off: International credibility and commitment to values against additional implementation and oversight burdens.
Feasibility: A sensible overarching frame for the sectoral approach. The real work lies in the implementing legislation.
A clear liability regime covers damage caused by autonomous AI decisions — for example, operator liability or hazard-based liability with an adjusted burden of proof, complemented by insurance solutions. For Switzerland, this closes the central gap where trust and insurability currently fail. It gives legal certainty to users and those who suffer harm alike. The task is to distribute responsibility fairly between manufacturer, operator and user without stifling innovation. The approach is demanding, but it addresses the actual structural break directly.
Trade-off: Legal certainty and protection for those harmed against the risk of stifling innovation through liability burdens.
Feasibility: The most effective lever against the structural break. It should complement the sectoral approach deliberately, not replace it.
Switzerland adopts a comprehensive, risk-based AI law analogous to the EU regulation, aiming for maximum compatibility with the European market. The advantage lies in easier market access and uniform rules across all sectors. The Federal Council has, however, deliberately decided against this path, viewing it as too rigid and poorly suited to Swiss regulatory practice. Feasibility is correspondingly low, even though de facto alignment through market pressure is likely to occur in part regardless.
Trade-off: Frictionless EU market access against a loss of flexibility and sovereignty.
Feasibility: Politically rejected, but important as a reference point, since market pressure will in any case drive partial alignment.
Trigger-based phases
Trigger: Federal Council decision on the regulatory approach in early 2025.
The technology-neutral, sectoral course is set and ratification of the Council of Europe AI Convention is under way. This phase produces consultation drafts and identifies the affected sectors, without any binding new obligations yet in force.
Trigger: Publication of consultation results and implementing legislation.
Individual sectoral laws are amended and the open question of liability for autonomous decisions addressed, supported by insurance solutions. In parallel, work is under way to assess where alignment with European standards is required for market access.
Trigger: First use cases and disputes arising in practice.
The sectoral rules are reviewed for gaps and inconsistencies and consolidated into a coherent overall framework. Case law and supervisory practice establish clarity on liability and accountability in day-to-day application.
Real-world pilot programmes
Status: In force
The Data Protection Act, in force since September 2023, regulates automated individual decisions, transparency and rights of access, and is enforced by the Federal Data Protection and Information Commissioner. It shows how AI-relevant obligations are already anchored in technology-neutral terms today.
Status: Ongoing
The financial markets regulator and the medicines authority are developing expectations for the use of AI in their respective domains — covering governance, explainability, and accountability. This sectoral practice is the concrete test of the chosen approach.
Status: Reference
The risk-based, horizontal EU regulation sets uniform obligations across all sectors and in practice also shapes Swiss providers with access to the EU market. For Switzerland, the question is which elements can usefully be incorporated without relinquishing its own sectoral approach and flexibility.
What we are watching
The measure tracks how far the announced consultation and the sectoral legislative adjustments have progressed. The milestone matters because only binding rules create legal certainty — without them, the decision of principle has no effect. A transition to enforceable law means the framework moves from intent into practice. For companies and investors, this is the point at which planning certainty begins.
The measure tracks whether clear attribution of responsibility exists for damage caused by autonomous AI decisions. The milestone matters because without it, insurability and trust are absent and widespread use remains constrained. A transition to clarified liability means the central structural break is closed. For those affected, insurers and users, this is the decisive measure of legal certainty.
The measure tracks how far the Swiss approach diverges from the EU regulation and whether that divergence hampers market access. The milestone matters because too great a gap forces Swiss providers to meet the EU standard regardless, which undermines the value of independence. A transition to a barrier means competitive disadvantages for exporting firms. For location policy and providers, this is the gauge of whether sovereignty and market access remain in balance.
The measure tracks the state of ratification and domestic implementation of the framework convention. The milestone matters because only completed implementation makes the values-based foundational principles legally binding. Finalisation signals international compatibility and a credible foundational framework. For policymakers and civil society, this is the signal that fundamental rights are durably anchored in the age of AI.