Swiss {ai} Weeks 2026 ran from September 1 to October 4, billed as the largest open movement for artificial intelligence and the official pre-event to the Road to AI Summit Geneva 2027. The Swiss Government is a listed partner, alongside the ETH and EPFL AI Centers, UBS, Swisscom and Swiss Re, according to the event site. The festival has now closed. The legal framework it celebrated has not yet been written.
What Bern has actually decided
On 12 February 2025 the Federal Council announced that it would ratify the Council of Europe's AI Convention and make the necessary changes to Swiss law. It also said that work on sector-specific AI regulation, in areas such as healthcare and transport, would continue. The official press release records both points. The Federal Chancellery's AI page says a draft bill for public consultation is expected by the end of 2026. The same date applies to an implementation plan for non-binding measures.
Law-firm analysis of the decision by Lenz & Staehelin describes three objectives. They are to strengthen Switzerland as a hub for innovation, to protect fundamental rights, and to build public trust in AI. The Council will not adopt a general cross-sector AI law. General rules are confined to fundamental-rights areas such as data protection. Everything else is left to existing sectoral regulators, supplemented by self-declaration agreements and industry solutions.
In practice, then, Switzerland is not planning an equivalent of the EU AI Act. Implementing the Convention will mean legal measures on transparency, data protection, non-discrimination and supervision. The draft is meant to come from the Federal Department of Justice and Police, with DETEC and the FDFA. The Federal Chancellery page adds that the Competence Network for Artificial Intelligence moved to the Chancellery in February 2026.
The strongest case for moving faster and harder
Critics of this approach make a serious argument. A sector-by-sector model can leave gaps where no regulator owns a use case, such as hiring tools, credit scoring or public-sector decision systems. Voluntary commitments are easiest to give when they cost nothing. Citizens harmed by an automated decision may find nobody with clear jurisdiction. A single horizontal statute would at least give them a defined entry point. There is also a practical point. Swiss firms selling into the EU will face the AI Act's requirements anyway. Some argue that a Swiss framework should therefore be written now so that Swiss companies have a say in how it develops.
These objections deserve a real answer, and the Council's own design gives part of one. The Convention work targets the areas where gaps matter most: transparency, non-discrimination, supervision, and the conduct of the state itself. The people who can least walk away from an automated decision are those subject to the state's. Starting with public-sector accountability is a proportionate place to begin.
Why the narrow path is still the better one
A horizontal AI statute fixes definitions and risk categories at a moment when the technology is changing quickly. Switzerland's existing law already reaches AI outputs through data protection, product liability, financial-market supervision and medical-device rules. Where a harm is real, a sectoral regulator that understands the market can act on it. Where it is speculative, a general statute tends to impose compliance cost on everyone and produce uncertainty that falls hardest on small firms and university spin-offs.
That matters in Switzerland because the AI Weeks partner list is a fair picture of the ecosystem. It includes two major university AI centers, large insurers and banks, and a start-up scene around Impact Hub Switzerland and Kickstart Innovation. Those are the actors a proportionate regime should serve. Bern's stated aim of protecting rights while preserving economic freedom is the right balance for them.
The risks sit in the details
The approach has three weak points that readers should watch.
- Timing. The Convention was signed in March 2025, and the consultation draft is due only by the end of 2026. A consultation, parliamentary debate and possible referendum period would follow. Convention obligations will not be in Swiss law for some time, and a pre-event for a 2027 summit will take place in a country that has promised rules but not yet shown them.
- Scope creep. The Council has said the Convention applies primarily to state actors, and the draft should hold to that. Extending binding duties to private developers through the implementing law would amount to a back-door AI Act, with its own cost.
- Soft law with no teeth or no tests. Self-declaration agreements are worth having only if their content is public and compliance can be checked. An implementation plan due the same month as the bill invites the two to be drafted in different directions.
What to watch
The consultation draft will show whether the Council keeps its own design. Four tests matter. Does it limit binding duties to public authorities and fundamental-rights areas? Does it name the regulator responsible in each sector? Does it publish the criteria for voluntary commitments? And does it arrive by the end of 2026? A narrow, on-time draft would show that sector-specific regulation is a deliberate choice and not a delay. A late or sprawling one would give critics the argument they are already making.
Switzerland has bet that trust and innovation can be built together without a flagship statute. That is a defensible bet and a more evidence-friendly one than prescribing rules for a technology still being invented. It will be judged on the draft, not on the events around it.