On 28 October 2026 the Swiss Federal Council did something governments rarely do with AI: it adopted nothing. According to the official press release, it "took note of the progress of work to date and the initial results" on implementing the Council of Europe AI Convention. It received two documents from the Federal Office of Communications (OFCOM/BAKOM). One is a report on sector-specific AI measures across the federal administration. The other is a guide for associations and firms that want to draft voluntary AI codes of conduct. The binding measures are still being drafted, and their consultation is now scheduled to begin in spring 2027.
The case for moving faster
The argument for urgency is serious and deserves a fair statement. AI systems are already used in hiring, credit, health triage and public administration. Voluntary codes depend on goodwill, and goodwill is thinnest where the stakes are highest. A binding floor on transparency, non-discrimination and supervision would give citizens a remedy and give firms a single reference point. Every quarter without one is a quarter in which harms can occur with no statutory recourse. Critics also note that a slipping timetable can become a habit.
That case is real. It is weaker for Switzerland than it first appears, for three reasons.
What Switzerland actually decided in 2025
The October step follows from a choice made on 12 February 2025. As the Federal Council's release put it, Switzerland "intends to ratify the Council of Europe Convention on Artificial Intelligence (AI) and to make the necessary amendments to Swiss law." It also said work would continue on AI rules "in specific sectors such as healthcare and transport." OFCOM's AI page describes the design: where legislative changes are needed they should be "as sector-specific as possible," and "only key areas relevant to fundamental rights, such as data protection, will be subject to general regulation." Legally non-binding measures are also envisaged.
This is a deliberate rejection of a single horizontal AI act. It is also the main reason the process takes longer. A sector-specific approach cannot be drafted in one stroke, because someone has to find out what existing law already covers.
The sector report is the useful part
The OFCOM report is based on a survey of all federal agencies. It gives an overview of current and planned measures in sectors including transport, energy, health and media. The release notes that high-tech sectors and those touching intellectual property, data processing, digital infrastructure and regulated markets face particular challenges. It also names concrete activity: the Federal Office of Culture's AI dialogue with the creative sector, which began in November 2025, a call by the Federal Office of Energy to the electricity sector, and OFCOM's discussions with the media sector.
This inventory is what a proportionate regime needs before it legislates. The EU's AI Act started from the opposite end, with a horizontal risk taxonomy that sectoral regulators then had to fit into existing product-safety and professional-conduct regimes. Switzerland is first asking which gaps actually exist. If the answer is that aviation, medical-device and financial rules already cover most of the risk, then new law should be narrow. Regulating only the gaps is a defensible reading of proportionality, and it is cheaper for small firms.
Voluntary codes: promising, but only with teeth
The Guide on an AI Code of Conduct is described on OFCOM's page as a toolkit that helps associations and businesses draw up voluntary guidelines for the use of AI. For a country of mostly small and medium-sized firms, industry-drafted codes can move faster than statute and adapt as technology changes. They also avoid locking in definitions of "AI" that may be obsolete within a few years.
The weakness is credibility. A code that nobody audits and nobody can enforce is a press release. Two safeguards would help. First, codes should state measurable commitments, such as disclosure when a person is interacting with an automated system, and a route to human review of consequential decisions. Second, the federal administration should say how code adherence will count in later supervision. Otherwise the voluntary track and the binding track will drift apart.
The scope question remains open
Law-firm analysis of the 2025 decision notes that public institutions must fully comply with the Convention, while private entities are only required to adhere to its fundamental principles. That analysis expected a consultation draft by the end of 2026. The new spring 2027 date therefore marks a slip of a few months, not a collapse. It also means private-sector obligations are the live question for the consultation. The press release lists transparency, data protection, non-discrimination, supervision and risk assessment as the subjects of the binding measures. Each of those can be written narrowly, as duties on high-impact uses, or broadly, as duties on any use of a model. A broad version would burden developers and deployers who pose little risk.
What to watch
Three tests will show whether the delay was worth it.
- Narrowness. The consultation draft should regulate identified gaps, not restate rules that already exist in sector law.
- Supervision design. Existing regulators should take on AI oversight within their mandates. A new AI authority should not be created by default.
- Honest timelines. The release sets a date of end-March 2028 for an internal knowledge-sharing platform inside DETEC. Public-facing milestones deserve the same precision.
A government that waits to understand its own legal landscape is behaving more responsibly than one that legislates first and amends later. But patience is only a virtue if the spring 2027 date holds and the voluntary tools are tied to something real. Switzerland has bought time, and the draft will show whether it used it well.