World-Class Research and Innovation

Frontier AI development is concentrated in a small number of private labs and two geopolitical blocs, leaving democratic societies without the public-interest infrastructure needed for independent safety research, evaluation, and pre-deployment testing.

A treaty-based, publicly funded CERN for AI would pool compute, talent, and evaluation infrastructure across democracies. Switzerland’s neutrality and its role as host to CERN, the ITU, and the UN Office at Geneva make it uniquely placed to convene such an institution

Deliver, by the Geneva AI Summit 2027, a scoped framework for a CERN for AI including mission, governance, funding, hosting, and a phased roadmap. Ready for endorsement by founding member states and institutions

1. An intergovernmental treaty modelled on the CERN Convention, with open accession for democratic states. 
   

2. Pooled public funding with a multiyear funding floor to guarantee independence from commercial interests. 
   

3. Democratic, multi-stakeholder governance. Council of member states, Scientific Council, and standing seats for civil society and independent safety experts.

4. Shared infrastructure: sovereign compute, evaluation testbeds, model and data access protocols, and secure facilities for dual-use work.

Primary Developers and operators of frontier AI models with systemic risk.

Secondary Governments, AI Safety Institutes, and researchers requiring trusted, independent access to evaluation infrastructure.

Public beneficiaries Citizens and democratic institutions exposed to systemic AI risks.

After years of "third-country" status, Switzerland was sidelined from the design of major European digital support streams. Success in these calls proves that Swiss institutions are not just "participants" but the architects of the European AI and compute landscape. The projects help to compensate lack of capital to scale solutions but also to increase integration into the EU startup and tech ecosystem.

Ensure that Swiss researchers and startups succeed in European digital R&I programs and fully use its potential still in 2026.

1. Formalize mechanisms to capture, analyze, and disseminate open EU AI support schemes.

2. Provide "triage" services for funding calls addressing sectoral AI adoption focusing on Swiss sector dominance.

3. Address more startups with these services to engage actors beyond the research ecosystem for EU R&I collaboration.

Scientific stakeholders and the startup ecosystem

A bottleneck in Swiss AI adoption is isolated innovation, where SMEs struggle individually with technical and regulatory barriers already solved elsewhere. EDIHs address this by acting as knowledge aggregators that pool cross-border expertise and proven solutions from over 200 European hubs.

In the course of 2027 develop Swiss EDIHs to one stop shops for AI SME support (legal, skills, tools) embedded in the AI Innovation ecosystem

1. Ecosystem Linkage: Orchestrates the connection between EDIHs and national/international initiatives, including AI compute facilities (e.g. ALPs), access to Data Spaces, Innovation Sandboxes

2. Shared technical assets (e.g., benchmarking software, legal and technical guidelines) are distributed and made comprehensible through the EDIHs.

3. Help SMEs overcome regulatory hurdles in the case of AI law, Data law, Cyber law

EDIH Governance Board

SMEs

Switzerland faces a critical gap in both technical and regulatory testing infrastructure, leaving AI stuck in "labs" and unprepared for real-world application. Without testing environments, startups lack the resources to scale, while established industries remain paralyzed by liability risks and the complexity of emerging digital laws.

Create terms of establishment and use for national AI regulatory and innovation sandboxes and define the legal requirements until end of 2026

1. Combine Technological Sandboxes (physical infrastructure/digital twins) and Regulatory Sandboxes (with authority oversight) under a single umbrella, while maintaining distinct organizational structures and participation terms.

2. Integrate sandbox approaches into hard law to provide clarity on regulatory treatment.

3. Enable testing with temporary exemptions to prevent innovation-stifling legislation for selected regulations in the digital field.

4. Establish a dual-track sandbox architecture by integrating horizontal (cross-sector) and sector-specific testing environments to ensure clear governance and jurisdictional alignment at the national level.

AI product developers

Both public institutions and private industry in Switzerland run on highly sensitive, regulated data, yet most generative-AI value today is captured by closed  models that route prompts and documents through external infrastructure. Apertus gives Switzerland a unique opportunity to build deeply embedded vertical solutions on a sovereign foundation: fully transparent, on-premise deployment, and verified-data connectors that keep regulated data within Switzerland.

Custom data and specialized derivatives close the gap to the deeply regulated workflows of key Swiss industries, delivering complete auditability, independence and full control, providing an instrument to manage dependencies.

Mobilize industry players and the public sector to build capabilities through joint learning. Generally enabling deployment on regulated sector tasks, while preserving full openness, on-premise deployability, Swiss data residency and auditability. Deliver replicable blueprints (datasets, fine-tuned checkpoints, evaluation suites) that are available for swiss key industries or public authority.

Hack Apertus
Hack Apertus creates replicable blueprints for AI adoption across industries and public administrations. Interdisciplinary academic and professional teams worldwide source and curate datasets across text, image, and voice, fine-tune Apertus 8B v1.5, and build open, on-premise use case solutions with public and private sector partners.

All outputs, including datasets, fine-tuned models, and Docker images, are published as open commons and possibly could be fed back to the Swiss AI Initiative as training data for next Apertus iterations.

Apertus for Finance
To build value in specific verticals, joint work beyond contribution from individual organizations is required. After workshops conducted by SNAI with over 40 financial industry players, a substantial overlap emerged in the need to identify use cases, benchmarks and make relevant training data available to be integrated into Apertus (apertvs.ai), while data upsampling of Apertus (1.5) is currently already ongoing in health, education and justice.

Pioneering financial industry players agree on relevant use cases, collaborate on supplying data and benchmarks, and pool required funding for the development of an Apertus for Finance. The Swiss Financial Innovation Lab, an independent Think & Do Tank (TDT) builds up the necessary ecosystem, helps win SNAI members and partners, coordinates company data inputs for potential Apertus training, but also guides its TDT members in maximizing output once the specialized model is developed.

Swiss FinTech Innovation Lab

Primary Swiss key industries (Piloted for Financial Sector) and the public sector