On 2 October 2026 the Swiss Federal Council adopted its report on raising the efficiency of administrative processes through process automation and artificial intelligence. It answers Postulate 24.3582 (Silberschmidt). The report is modest in tone, and that suits Switzerland's approach. But it also leaves open how anyone will know whether it works.
What the Federal Council actually decided
The Federal Council's announcement sets five priority areas for AI and automation in the federal administration:
- business administration, such as parliamentary business and inter-agency consultations
- review of applications and cases
- technology-assisted translation
- data management
- operational support processes, such as finance and procurement transactions
It says deployment must be "human-centred, trustworthy and sustainable". The stated motivation is demographic. The administration must handle equal or growing workloads, and rising data volumes, with fewer staff. The next steps are better portfolio tracking of existing AI and automation systems, identifying synergies, prioritising high-impact use cases and scaling what works under the existing federal AI strategy.
The announcement gives no budget, headcount or expected efficiency gain. That matters for the argument below.
The case for caution, stated fairly
The strongest argument for a heavier hand is that government use of AI differs from private use. Citizens cannot opt out of a benefits decision or a permit review. Automated application processing can embed errors at scale, and an opaque system makes appeal harder. A government that deploys AI without binding rules on transparency, human review and non-discrimination risks eroding the trust it says it wants. That is why many advocates want legal duties that apply to the state first.
Switzerland has not ignored this. The Federal Chancellery says the administration already has an AI strategy with eight implementation measures, general guidelines, and a Competence Network for Artificial Intelligence. The Chancellery's DTI Sector has run that network's contact point since 1 February 2026. The new report sits on top of this existing structure.
Why the report's approach is right in direction
The five priority areas are well chosen. Translation, application processing and back-office work are bounded, repetitive tasks. They are also where the burden falls on a multilingual state that works in German, French and Italian. They do not involve the discretionary, rights-affecting decisions that raise the hardest legal questions. Starting where the error cost is lowest and the volume is highest is proportionate. It lets the administration build competence before it touches contested ground.
The decision to track existing solutions and scale successful ones is also sound. Governments often fail by launching many pilots that never reach production. A portfolio view, with an explicit duty to scale winners, targets that failure directly.
This fits the stance the Federal Council took on regulating the wider economy. On 12 February 2025 it decided to ratify the Council of Europe's AI Convention and implement it through sector-specific amendments, rather than a single horizontal AI act. Its stated objectives are to strengthen Switzerland as a location for innovation, to protect fundamental rights including economic freedom, and to increase public trust in AI. Non-binding measures such as industry agreements and self-declarations are part of the mix. For a small, export-oriented economy this is a defensible alternative to the EU's AI Act model. It keeps rules close to the specific harm, in healthcare, transport and elsewhere, instead of regulating a general-purpose technology as a category.
Where the report falls short
The weakness is measurement. A strategy that promises to "scale successful solutions" needs a definition of success. The announcement names none, and it offers no baseline of how many AI systems the administration now runs. It sets no target for processing times, cost or staff hours, and no schedule for publishing results. Without these, parliament and the public cannot tell whether the programme is delivering. The postulate's own premise was efficiency, and efficiency is a quantity.
There is also a sequencing issue. The Federal Department of Justice and Police, working with two other departments, is due to deliver a consultation draft implementing the Council of Europe Convention by the end of 2026. That draft is expected to cover transparency, data protection, non-discrimination and supervision. The government is therefore building its own AI use before the legal rules that will bind it are public. This need not be a problem, but the administration should treat the coming rules as constraints from the start. It should avoid building systems that must later be rebuilt.
The same news coverage records the political divide. The Greens called the approach insufficiently ambitious, while the union Syndicom backed the framework but wanted mandatory risk assessments extended to private companies. The government can answer both camps by showing its own use of AI is transparent. A public register of federal AI systems, each with a stated purpose, a human-review point and an appeal route, would cost little. It would make the "trustworthy" commitment checkable, and it would give the later legislative debate evidence to work from.
What to watch
Three things will show whether this report is substance or signalling:
- Metrics: whether the federal AI strategy publishes baselines and targets.
- Transparency: whether the portfolio tracking becomes public or stays internal.
- Alignment: whether the end-2026 consultation draft treats the state's own AI use under the same transparency and non-discrimination duties as private actors.
Switzerland's light-touch, sectoral model is a credible way to pair innovation with rights protection. A light touch still has to be accountable. The Federal Council has chosen sensible places to start. To earn trust it now needs to say what success looks like, and report against it.