Internet Archive Switzerland launches in St. Gallen

On May 5 2026 the Internet Archive announced the opening of a Swiss branch in St. Gallen. The foundation positions itself as a non‑profit dedicated to “universal access to all knowledge,” echoing the mission of its U.S., Canadian, and European counterparts. Two flagship initiatives were highlighted:

1. Gen AI Archive – a collaboration with the University of St. Gallen to preserve today’s generative‑AI models and related datasets.
2. Endangered Archives – a program, in cooperation with UNESCO and other partners, to rescue digital collections threatened by conflict, disaster, or censorship.

---

What is being claimed?

• Comprehensive AI preservation – the foundation claims it will become “the world’s first comprehensive AI archive,” storing model weights, training data, and documentation for future research.
• Secure haven for at‑risk collections – the Endangered Archives initiative promises a “digital haven” where vulnerable cultural heritage can be stored, replicated, and made publicly accessible.
• Alignment with UN human‑rights goals – the charter cites Articles 19, 26, 27 of the Universal Declaration of Human Rights and Sustainable Development Goal 4 as guiding principles.

What is actually new?

A dedicated Swiss legal entity

The most concrete novelty is the creation of a Swiss legal entity (registered as CHE‑238.343.099). This gives the Internet Archive a local governance structure, a board chaired by Executive Director Roman Griesfelder, and a physical address in St. Gallen. The Swiss foundation model provides tax‑exempt status and a clear framework for fundraising within Europe.

Partnership with the University of St. Gallen

The Gen AI Archive is not merely a marketing tagline; the university’s Department of Business Informatics has already begun cataloguing three open‑source language models released in 2023‑2024. The partnership includes:

• A shared storage cluster hosted at the university’s data center, currently 200 TB of capacity.
• A joint advisory board that will define metadata standards for model provenance (e.g., version, training corpus, licensing).
• An initial pilot to ingest the weights of the LLaMA‑2‑13B model and its associated training logs.

You can read the university’s brief on the project here.

Endangered Archives pilot with UNESCO

The foundation has signed a memorandum of understanding with UNESCO’s Memory of the World Programme. The first test case will involve digitising a collection of newspaper microfilms from a war‑torn region in the Balkans. The workflow mirrors the existing Internet Archive’s “Wayback Machine” ingestion pipeline but adds a checksum‑based integrity check every 30 days.

More details are available in UNESCO’s announcement PDF.

Limitations and open questions

Sustainability of storage costs

Preserving large AI models is expensive. A single 175‑billion‑parameter model can exceed 700 GB in compressed form. Even with the university’s 200 TB cluster, the foundation will quickly outgrow its initial allocation. The announcement does not specify a long‑term funding model beyond typical grant applications.

Legal and licensing hurdles

Many state‑of‑the‑art models are released under restrictive licenses (e.g., non‑commercial clauses). The Gen AI Archive will need to navigate copyright and export‑control regulations before making any model publicly downloadable. The current plan mentions “controlled access” but provides no technical details on how access will be gated.

Metadata standards are still evolving

The AI community has not converged on a universal schema for model provenance. The foundation’s advisory board will have to decide between emerging standards such as the Model Card framework, the MLflow tracking format, or a custom schema. Until a consensus is reached, the archive may end up with heterogeneous records that are hard to search.

Risk of “digital dark age” persists

While the Endangered Archives initiative aims to rescue at‑risk collections, the process still relies on external partners to provide source material. In conflict zones, getting physical media to a digitisation lab can be impossible, and the archive’s ability to act quickly is limited by logistics, not technology.

Practical implications for researchers

• Access to historic AI models – If the Gen AI Archive succeeds, scholars will be able to reproduce experiments from 2023‑2025 without recreating the original training environment, a common pain point in reproducibility studies.
• Preservation of cultural heritage – The Endangered Archives pilot could become a model for other NGOs seeking a “last‑ditch” backup for collections at risk of destruction.
• Potential for collaborative standards – By publishing its metadata schema, the foundation could influence broader community adoption, reducing fragmentation across AI model repositories.

Bottom line

Internet Archive Switzerland brings a formal Swiss presence to the global preservation effort, and its two flagship projects address real gaps: the long‑term storage of generative‑AI artifacts and the rescue of vulnerable cultural collections. The initiative’s success will depend on securing sustainable funding, clarifying licensing pathways, and establishing interoperable metadata standards. Until those pieces fall into place, the foundation remains an ambitious but still‑nascent addition to the preservation ecosystem.

---

For more information about the foundation’s governance and contact details, visit the official site Internet Archive Switzerland.