Twenty-two indications examine cloud, data, software, work, rights and consumption, while the topic also enters the federal debate

Don't adopt AI because it's available, but decide first what public problems it needs to solve, which activities can be automated and which technological dependencies are acceptable. This is the criterion around which Geneva is building part of its digital policy, after commissioning an interdisciplinary study on AI and technological autonomy from the University of Geneva. The work formulates 22 recommendations ranging from public employee training to the cloud, from open source tools to procurement, all the way to fundamental rights and the ecological footprint of models.
The study was conducted by an interdisciplinary group from the University of Geneva composed of Cedric Durand, of the Département d'histoire, économie et société of the Faculté des sciences de la société, Yaniv Benhamou, of the Département de droit commercial of the Faculté de droit, Diego Kuonen, of the Faculté d'économie et de management, e Gaia Valenti, of the Faculty of Social Sciences. The mandate comes from the Department of Institutions and Numerics (DIN), while the document is intended to inform the future cantonal strategy on artificial intelligence.
Therefore, these are not 22 measures already approved: some coincide with ongoing construction projects, while others will need to be evaluated in terms of feasibility, federal and intercantonal competences, legislation, and financial resources. Taken together, the recommendations can be traced back to three major orientations: govern the pace and extent of AI deployment in government, maintain control over infrastructure, data, and technology dependencies, and protect people while also considering the systems' environmental impact.
The starting point is a pragmatic definition of digital sovereigntyThe study does not identify it with technological autarchy, considered unrealistic, but with the capacity of public authorities to choose and manage your own interdependencies and to remain in control of their own decisions even as technologies, platforms, and the balance of power between suppliers change. Hardware, computing capacity, cloud, software, data, and models form a chain in which the autonomy of a single component is not enough: a public entity can physically store data on its own territory and still be tied to platforms, formats, services, or suppliers that are difficult to replace.

First the public objectives, then the choice of AI tools
The first five recommendations address the speed with which AI enters public services. The authors ask to first define the political and administrative objectives and the concrete needs of the different professions, avoiding that organizational transformation is driven simply by the possibilities offered by technology. This principle is accompanied by continuous programs of data literacy, interdisciplinary training in high schools and universities and consultation with staff to establish professional standards on the use of automated systems.
The issue isn't just about mastering new software. The introduction of AI could change the distribution of skills within offices, the responsibility for decisions, and employees' ability to recognize an incorrect outcome. The document specifically highlights the risks of professional deskilling, uncertain reliability, deterioration of cooperation and loss of internal knowledge when cognitive functions are transferred too quickly to machines. Preserving skills therefore becomes part of the strategy: the authors also consider automation as an organizational and management issue. digital transformation, not just computer science.
This also leads to the call to involve staff and share experiences and solutions more widely among administrations. Limited experimentation should precede the extension of a system to entire processes, allowing for the verification of benefits, errors, costs, and consequences on work before making a technology structural. Geneva also indicates that the guiding principle should be to develop AI applications when they produce a actual added value, while maintaining control over the conditions of their use.
Cloud, open source and procurement become tools for autonomy
The most industrial part enters the artificial intelligence technology stackThe sixth to ninth indications propose to give priority, in the public sector, open source and frugal AI models, accelerate the transition towards a private cloud under public control, promote the pooling of resources and solutions among public entities, progressively migrate to open source software, and limit the lock-in effects associated with proprietary platforms. They also recommend prudent data management and contractual clauses that guarantee their accessibility to entities.
Lock-in is one of the concrete problems of digital autonomy: it arises when the technical, economic, or organizational costs of abandoning a supplier become so high that they reduce the customer's freedom of choice. For a public entity, this may concern data compatibility, application interfaces, staff skills, cloud architecture, or applications built around the services of a single operator. The issue therefore intersects both security and privacy as well as operational continuity and bargaining power.
The authors also assign a role to public procurementOne proposal is to use institutional procurement to enable local operators to develop expertise across the various layers of the stack, from applications to the cloud. Another suggests making major public aid to businesses, including startups, conditional on precautions against infrastructure dependence on suppliers from a single country.
The thirteenth proposed measure is more incisive: to take public shareholdings or to assign Geneva a golden share in local digital companies considered strategic, thus having veto power over decisions affecting vital interests. The study also raises the possibility of limiting foreign investment in companies considered critical infrastructure. This is a proposal by the authors, not a decision already made by the Geneva authorities, and is one of the points that would require legal, economic, and institutional review before any implementation.
The approach shows how the technological control It's not just about server location. It includes infrastructure management, the ability to change software, data availability, the expertise needed to manage systems, and the presence of multiple suppliers capable of competing. From this perspective, mutualization between agencies and procurement policies can change the very structure of the digital ecosystem with which the public sector interacts, fostering greater diversification and resilience.
Automated decisions, appeals and transparency on technical systems
Points seventeen through nineteen shift the focus to individuals. The authors ask to verify whether cantonal law adequately protects individuals. fundamental rights in the use and development of AI and, if not, to evaluate new regulations. The creation of a Cantonal register of AI-based decision-making systems and periodic risk self-assessment reports.
The registry would make it possible to identify the areas in which an algorithm participates in processes capable of impacting people. The issue becomes particularly sensitive when AI intervenes in public services, evaluations, classifications, or administrative procedures. Transparency is therefore accompanied by the demand for effective appeal procedures, with tools that allow the effects of a decision to be contested and, when the conditions exist, responsibilities and sanctions to be identified.
Geneva already has a unique reference: the right to digital integrity It is enshrined in the cantonal constitution. At the same time, the AI guidelines are being updated to clarify responsibilities and employment conditions in the public sector. The strategy approved in June 2026 constitutes the broader framework encompassing infrastructure, data, software, expertise, and supplier relationships.
This drawing also includes the proposal to create a Interdisciplinary Scientific Council for Digital at the DIN, or at a broader institutional level that can provide support to the authorities. The document also suggests supporting Geneva's international role in AI regulation and the cooperative development of open, resource-efficient technologies.
Energy, water and raw materials also enter the AI budget
The last three proposals address a less visible aspect of the artificial intelligence: its material cost. The researchers propose to make the data relating to the electricity consumption and ecological impact associate with AI in the public and private sectors, set multi-year targets to reduce the footprint and support more frugal and specialized models.
Energy, water, land for data centers, and materials needed for hardware thus become parameters of digital policy along with performance, security, and costs. Recommendation number 22 explicitly links research and development to technologies that are less demanding on material and energy resources, including more sustainable practices for data collection and storage and for model training.
The Geneva study comes as the issue is also taking on a federal dimension. Swiss Digital Pact he welcomed it as an element consistent with the theses of his popular initiative “For Switzerland's digital security”The association argues that uniform principles at the constitutional level can prevent fragmented cantonal approaches and calls for the Confederation to be given more explicit powers over security, infrastructure, and the country's ability to maintain digital decision-making margins. This is the position of the initiative's promoters, distinct from that of the Geneva authorities.
The text of the initiative proposes a new Article 57a of the Federal ConstitutionThe provisions include security rules for public and private actors in the Swiss digital space, protection of federal data and infrastructure, protection of personal data and digital integrity, safeguarding essential digital resources from influences contrary to the country's interests, and promotion of digital skills. The signature collection began on June 2, 2026; for the formal success of a federal popular initiative, it is necessary 100.000 valid signatures and, for this proposal, the deadline indicated is December 2, 2027.
The work commissioned by Geneva doesn't determine the outcome of that debate and doesn't equate to cantonal support for the initiative. However, it offers a concrete example of how technological autonomy can translate into operational choices: deciding where to store data, which dependencies to accept, which skills to retain internally, how to write tenders, when to make an algorithm transparent, and what resource consumption is considered sustainable. The 22 recommendations will now be incorporated into the reflection on the future cantonal strategy for artificial intelligence: some may build on existing initiatives, while others will require feasibility and institutional, legal, and financial compatibility assessments. The stated objective remains to allow Geneva to exploit the opportunities of AI without sacrificing its control over one's own technological choices and public policies.
Here are three insights that might interest you:
Apertus is the Swiss "open AI" for a more transparent future.
For Tim Berners-Lee in the age of AI the Web must be given back to the people
Swiss {ai} Weeks Bring Artificial Intelligence to 37 Cities











