A plan without a law behind it — yet
On May 26, 2026, roughly 200 executives, researchers, and officials gathered in Andermatt for the Digital Gipfel Schweiz, where digitalswitzerland — the country's umbrella digitalisation association — launched the AI Action Plan for Switzerland in coordination with the Federal Office of Communications (OFCOM/BAKOM) (digitalswitzerland; Bilanz). The plan sets out 23 concrete actions across five pillars: scaled AI literacy and education, world-class research and innovation, resilient digital infrastructure, AI-ready data, and smart AI governance. digitalswitzerland CEO Franziska Barmettler summed up the mood: "The consensus of this year's Digital Gipfel is clear: we must act decisively now."
What makes the plan notable is what it isn't. It is not legislation, and Switzerland does not yet have an AI law. The Federal Council decided on February 12, 2025 to ratify the Council of Europe's Framework Convention on AI and implement it through sector-specific adjustments to existing statutes rather than a horizontal EU-style AI Act, built around three objectives: strengthening Switzerland as an innovation hub, safeguarding fundamental rights, and building public trust (Federal Chancellery; OFCOM). A consultation draft is due only by the end of 2026, with actual legislation further out. The Action Plan is designed to fill that gap — an industry-and-regulator-coordinated roadmap operating in the two-plus years before any binding rule takes effect.
The steelman for going first with soft law
There's a real case for this sequencing. Switzerland is a small, open economy with no domestic hyperscaler and a research base — ETH Zurich, EPFL — that punches well above the country's population. If Bern spent the next two years drafting comprehensive AI legislation before any coordinated literacy, infrastructure, or data effort began, the country would fall further behind in exactly the capacities that determine whether AI regulation has anything real to bite on. A literacy campaign, open Swiss-language models for critical industries, and shared data infrastructure are prerequisites for a competitive AI sector regardless of what the eventual law says — and none of them require a statute to start. Coordinating them now, under OFCOM's involvement, also means the eventual legislative draft won't be written in a vacuum; it can draw on a year or two of live implementation experience about where voluntary commitments held and where they didn't.
The plan's most concrete deliverable illustrates the model: a national AI-literacy campaign launching in 2027 aiming to train one million people in responsible AI use, backed by an SME-focused "Playbook for AI Adoption" and AI certification tracks for public-sector roles (AI Action Plan). None of that needed to wait for a statute.
Where the soft-law bet gets risky
The weak point is the governance pillar itself. "Smart AI governance" in the plan's own framing means an "innovation-friendly, lean regulatory approach" supplemented by voluntary industry self-commitments — not enforceable rules. digitalswitzerland's own contributor page for the plan invites input from "business, science, and civil society" through a consultation platform, but publishes no detail on how contributions are weighted, who sits on the steering group beyond a list of corporate and academic names (EPFL, ETH AI Center, Accenture, IBM Switzerland, ELCA, Kudelski), or what happens if an "Action Contributor" simply doesn't deliver (digitalswitzerland). A voluntary roadmap with no enforcement mechanism is only as durable as the goodwill of the companies that signed onto it — fine for a literacy campaign, less fine if "AI-ready data" ends up meaning looser de facto standards for sharing personal or health data ahead of the Federal Council's own promised cross-sectoral rules on transparency, non-discrimination, and oversight.
There is also a sequencing risk the plan doesn't fully address: by the time Switzerland's actual AI statute reaches consultation at the end of 2026, an entire ecosystem of infrastructure and data-sharing practice will already have hardened around the Action Plan's voluntary norms. That's an argument for OFCOM staying closely embedded in the plan's governance workstream now — which, per BAKOM's own AI strategy page, is the stated intent — rather than treating the Action Plan and the eventual statute as separate tracks that only meet at the end.
The right comparison isn't Brussels, it's Washington
Switzerland is explicitly not copying the EU AI Act's ex-ante, risk-tiered licensing model, and it shouldn't: the compliance burden a horizontal act imposes on Swiss SMEs — most of Switzerland's economy — is disproportionate to the country's much smaller AI-safety exposure. The closer analogue is the United States' pre-statute reliance on voluntary commitments and sectoral enforcement, which has produced faster deployment with the tradeoff of thinner federal guardrails. Switzerland's version has one advantage the U.S. approach lacks: a regulator, OFCOM, sitting inside the voluntary structure from day one rather than watching from outside. If that involvement stays substantive rather than symbolic through 2027, the sequencing — infrastructure and literacy first, binding law second — is defensible. If OFCOM's role turns out to be limited to co-branding a press release, Switzerland will have spent two years building AI capacity on foundations nobody can enforce.