AI in government: six core measures to organise over 100 projects
The federal government is continuing its AI strategy with fewer funds. Six instead of eight measures are to coordinate more than 100 ongoing projects – including an internal generative system for employees.
The federal administration is continuing to implement its AI strategy, though with tighter priorities. On 18 September the Federal Council was informed about a new roadmap for 2026 to 2028: six of the original eight supporting measures remain. According to the Federal Chancellery, two million francs a year are available for them.
That sounds like an administrative detail, but it touches on a central question: how can the federal government use artificial intelligence without every office managing its own rules, tools and risks? According to the Federal Chancellery, 116 AI projects are already running in the departments and in the Federal Chancellery.
From pollen forecasts to confidential texts
The list of ongoing projects shows the range. It includes AI-supported forecasts for pollen levels, the detection of photovoltaic installations in aerial images and a writing aid for confidential texts. The projects have different purposes and different protection needs. A pollen model works with different data and risks than an application that processes administrative texts.
That is precisely why the Federal Chancellery wants to coordinate activities more closely. The strategy relies on trustworthy AI use, skills development and process optimisation. Whether that succeeds depends less on the number of pilot projects than on common standards, traceable responsibilities and the question of which data may be processed in which system.
Six measures remain
One of the prioritised points is a federal-wide survey of where AI could improve processes. A generative AI system for employees of the federal administration is also to be continued. Added to that are foundational documents for the legal framework and the AI competence network, which is meant to pool knowledge and experience between offices.
At the same time, the federal government is dropping parts of the original plan. A planned AI handbook will not be produced. The planned central AI marketplace, where offices would have been able to find and reuse solutions, is being replaced by an expansion of the existing project database. That is not merely a change of wording: a database can create visibility, but it does not automatically replace advice, review and common technical standards.
Fewer funds, a clearer order
The re-prioritisation is linked to the generally slowed pace of the federal government's digitalisation projects. Two million francs a year are not enough to support every idea centrally. The Federal Chancellery is therefore relying on measures that are meant to help several offices at the same time.
There is a sensible idea in that. Many authorities face similar questions when it comes to AI: Which legal basis applies? How are models controlled? How do you prevent confidential information from reaching external systems without authorisation? If these questions are answered separately, duplication arises.
At the same time, an internal solution must not be understood as a shortcut. A generative system for the federal administration needs clear rules of use, technical safeguards and transparency about where its limits lie. The announcement names the direction, but not the concrete features of a future system. Statements about the scope of functions, data processing or a launch date would therefore be premature.
The test comes by the end of 2028
The Federal Chancellery intends to publish the implementation on an ongoing basis; the measures are to be reviewed by the end of 2028 at the latest. Until then, the decisive question is whether the projects that are scattered today produce experience that can be reused.
The new roadmap is not a withdrawal of the federal government from AI. It is a concentration under financial constraints. Its success will not be measured by the number of 116 projects, but by whether authorities use AI in a legally secure and traceable way, and where it actually enables a better public service.
What remains open on oversight
The roadmap describes structures and priorities, but does not yet name uniform criteria by which the benefits and risks of individual projects are compared. Reliable oversight requires more than a list of ongoing projects: it should be recognisable which problem an application solves, who is responsible for it and how errors or unwanted consequences are measured. Traceable limits are crucial, especially for systems that process internal administrative or confidential texts. The announced ongoing publication of the implementation can create a basis for this. Whether it is sufficient depends on how concretely the federal government documents progress, abandoned projects and lessons learned. Without this information, the effectiveness of the new prioritisation cannot yet be conclusively assessed.
Sources
- Federal Chancellery: Federal administration prioritises AI strategy measures
- Federal Chancellery: Ongoing AI projects of the federal government
- Federal Chancellery: AI strategy of the federal administration
