At the Digital Gipfel Schweiz summit in Andermatt on 26–27 May 2026, Switzerland's national AI strategy crossed a quiet but consequential threshold: the AI Action Plan for Switzerland reached its "First Draft" milestone. Convened by digitalswitzerland with the Federal Office of Communications (BAKOM) and Accenture, the plan is deliberately not a statute. It is a rolling, public-private agenda — 23 concrete actions across five focus areas — that aligns federal leadership with industry and research institutions to keep Switzerland competitive in AI. The bet is that coordination, not a comprehensive AI Act, is the right instrument for a small, open economy.
What the plan actually is
The official action plan organizes its 23 actions under five headings: scaled AI education and literacy, world-class research and innovation through European cooperation, resilient digital infrastructure, AI-ready data, and "smart AI governance" — explicitly framed as innovation-friendly and streamlined. Crucially, these are commitments owned by named actors — companies, universities, and public bodies — rather than obligations imposed on a regulated population. digitalswitzerland describes it as "less a rigid strategy document and more an ongoing process," with an AI summit in Geneva in June 2027 set as the next culmination point.
The convening power on display is real. Switzerland's digital agenda at Davos 2026 drew Microsoft Switzerland, Google, Cloudflare, ETH Zurich, the EPFL AI Center, the National Cyber Security Centre, and the Council of Europe to the same table — pegging the prize at roughly CHF 15 billion in annual value from AI applications. That mix of frontier labs, hyperscalers, and two of Europe's strongest technical universities is precisely the asset a coordination model is meant to mobilize.
The steelman for a binding AI Act
The case against Switzerland's light-touch route deserves a fair hearing. A comprehensive, horizontally binding law — the EU's AI Act being the obvious template — delivers legal certainty, uniform fundamental-rights protections, and a single enforcement address that voluntary commitments cannot match. Pledges by firms are reversible; statutory duties are not. For high-stakes uses such as biometric identification, hiring, or medical devices, a baseline of enforceable transparency and oversight is a genuine public good, and the Council of Europe's own Framework Convention exists precisely because soft assurances have a poor track record under commercial pressure.
Why coordination is still the proportionate call here
Switzerland is not choosing between regulation and a vacuum. On 12 February 2025 the Federal Council decided on a sector-specific approach, opting to integrate AI into existing law rather than build a standalone regime, with three stated objectives: strengthening Switzerland as an innovation location, safeguarding fundamental rights and economic freedom, and increasing public trust. Switzerland signed the Council of Europe Framework Convention on AI in Strasbourg on 27 March 2025, and the relevant federal offices are to deliver a consultation draft implementing it by the end of 2026. So the binding floor — transparency, data protection, non-discrimination, supervision — is being built in parallel; the Action Plan sits on top of it as the growth engine.
This layering is the proportionate-regulation thesis in practice. A horizontal AI Act front-loads compliance cost onto every developer regardless of risk, and its categories ossify just as the technology shifts. A small economy that imports most of its model capability gains little by duplicating Brussels' rulebook and much by being faster, clearer, and friendlier to the researchers and startups it actually hosts. As Sidley's analysts note, the lean path is a genuine competitive edge — if executed well.
The risks the plan must not paper over
That conditional matters. Two failure modes are real. First, voluntary under-delivery: a rolling agenda of self-owned commitments can quietly slip when budgets tighten, with no enforcement to call it back. The plan's credibility will live or die on whether its 23 actions ship measurable outcomes by the 2027 Geneva summit, not on the elegance of its consultation process. Second, the "Swiss Finish" problem — divergence from EU standards that, instead of cutting red tape, adds a second incompatible layer for firms operating across the border, especially in life sciences and medical devices. Coordination that produces a bespoke Swiss interpretation of every EU norm would deliver the costs of fragmentation without the benefits of either model.
The verdict
Switzerland's First Draft is a defensible, even instructive, model: a binding rights floor anchored in an international treaty, paired with an industry-and-academia coordination layer to drive competitiveness, and no premature horizontal AI Act to tax innovation before harms are demonstrated. It is the proportionate posture this publication favors. But coordination is a promise, not a result. The test is execution — concrete, audited deliverables and tight interoperability with EU rules. If Andermatt's draft becomes Geneva's track record, Switzerland will have shown small jurisdictions a smarter path than copying the AI Act wholesale.