AI-Ready Data

Data play a critical role in running and training AI systems (ML, GenAI, agentic AI, etc.), but there is still limited awareness of its impact on generating trustworthy, reliable and responsible AI outputs.

Many data providers from the public and private sectors would like to make their data available for AI systems, but are unsure how to address the emerging AI requirements.

The developers of AI systems face challenges in getting data to build reliable, explainable, and scalable AI applications.

Define and create a shared understanding around AI-ready data.

Specify how datasets can be assessed with regards to their AI-readiness.

Develop guidelines for provision and consumption of AI-ready data

• Foundations of AI-ready data (Definition, dimensions, enabler, etc.) - Whitepaper, until October 2026

• Assessment of AI-ready data – Checklist, in 2027

• Target group-specific recommendationsfor AI-ready data - User guides, in 2027

  • Data providers: How can existing/curated data be prepared to support different types of AI, specifically GenAI  and agentic AI?

  • Data consumers: How can data be assessed and verified to ensure that AI systems produce trustworthy, reliable and responsible outputs?

General public (as users of AI systems)

Data and AI practitioners

Organizations /as developers and users of AI systems), specifically those which are “too small” regarding necessary technical resources and skills (e.g. SMEs).

The effective and scalable use of AI depends on the ability to responsibly reuse existing data beyond its original context of collection. However, secondary use of data is often limited by fragmented governance, unclear responsibilities, insufficient interoperability, legal uncertainty, and lack of trustworthiness. Promoting the responsible secondary use of data helps unlock existing data potential, strengthens innovation and collaboration, reduces duplication of effort, and supports the development of trustworthy AI systems across sectors.

Enable and promote the responsible, secure, and efficient secondary use of data for training, testing, operating, and improving AI systems across business, research, education, and the public sector.

• Common language and definition of secondary use of data for AI.
  

• Legal framework for enabling secondary use of data, specifically for AI systems – Rahmengesetz für die Sekundärnutzung von Daten
  

• Specific requirements for secondary use of data for AI: Standardised metadata and interoperability approaches, mechanisms for secure data access and sharing, etc.
  

• Support for data-sharing ecosystems and partnerships promoting the secondary use of data in critical domains (mobility, health, research, etc.)

Expert Group on the Framework Law for the Secondary Use of Data (tbc)

• Public sector

• Businesses

• Research and education

• Data and AI practitioners

• Policymakers and ecosystem partners

Public data including open data and administrative data are key resources for AI systems. However, Switzerland’s structure is fragmented, which makes it difficult to effectively and consistently provide access to up to date, high quality, and linkable data across federal, cantonal, and municipal levels.

 Lacking standards and interoperability on both metadata and data level will slow down data reuse and affect AI adoption in Switzerland.

Switzerland should extend and adapt existing standards and platforms to support AI, rather than “reinventing” the wheel.

• Define metadata standards that improve use of up to date, high quality, and linkable data in AI systems, starting with a core “minimal viable metadata standard” (e.g., per sector/industry)

• Specify standards to harmonize data (semantics)

• Extend (data) platforms to support AI systems

• Ensure binding standard setting competencies on federal level

• Extension of metadata standards for AI systems (i.e. move from static to dynamic machine-actionable metadata, considering critical elements for AI systems, e.g. versioning, time stamps)

• Data harmonization (semantics) for critical datasets / fields - develop classification criteria, ontologies or schema in a community-driven approach

• Extension of data platforms– focus on accessibility and Interoperability of platforms across federal, cantonal, and municipal levels.

• Public sector

• Research and education

• Policymakers and ecosystem partners

In an era of information overload and AI-generated content, discovering reliable, trustworthy data and statistics has become increasingly difficult. Even high-quality data produced by governments and international organisations is often invisible to search engines and AI tools. This contributes to confusion in the digital information space, where both accurate information and misleading or deliberately fabricated content – including disinformation and so-called “fake news” – can spread rapidly, hindering informed decision-making, weakening trust, and at times even obstructing international cooperation.

The Trusted Data Observatory (TDO) is a global metadata platform that solves the challenge of data discovery in an AI-enabled world. Instead of hosting data, TDO describes trusted data using a Minimum Viable Metadata (MVM) model that is standardised, open, machine-readable, and optimised for both humans and AI systems.

• Phase 1 (2025-2026; completed): Stakeholder identification, governance planning, and requirements definition have commenced in close cooperation with international organisations and national statistical offices.

• Phase 2 (2025-2026; completed): Agreement on a Minimum Viable Metadata (MVM) set, associated standards, technical specifications, and Proof of Concept (PoC) participants.

• Phase 3 (end 2026-Q1 2027): Development of a scalable TDO prototype, implementation of the PoC, training of participants, and engagement with AI and technology stakeholders.

• Phase 4 (by end 2027): Review of PoC, review and refine the TDO post PoC exercise.

• Phase 5 (2027-2030): Broader onboarding, integration with post-2030 agenda, and long-term sustainability.

• Initial focus on the data holdings of National Statistical Offices and International Statistical Organisation or International Organisations with dedicated statistical divisions.

• As the TDO evolves over time broader scope can be explored and incorporated.
  

The effective and trustworthy use of AI depends on the availability, visibility, and responsible reuse of AI-ready data across sectors. However, many datasets and use cases remain fragmented, insufficiently documented or not easily discoverable. This limits reuse, collaboration, learning, and scaling potential.

Promote the documentation and showcasing of relevant use cases, datasets, and practices related to AI-ready data across business, research, education, and the public sector in order to promote reuse, collaboration, learning, and responsible AI adoption.

• Documentation model for AI-ready data use cases: template (with datasets and application contexts)

• Collection of AI-ready data use cases (e.g. classification by sector, domain and maturity level) – in collaboration with data initiatives in domains

• Support for discoverability, reuse and exchange of good practices.

• Analysis and reporting of use cases

• Public sector

• Businesses

• Research and education

• Data and AI practitioners

• Policymakers and ecosystem partners