Public AI, Built Together — Towards Geneva 2027

A public library introducing AI, a municipality using it to prepare decisions, and a small organisation working in a less widely served language face related questions. Can they choose a suitable system, understand its limits, protect the information entrusted to them, and challenge an answer that affects someone? Can they retain those abilities when a provider changes its terms or withdraws a service?

Public AI should give people and institutions the capacity to answer those questions in practice. That requires models and computing resources, but also skills, shared evidence, accountable institutions and ways to cooperate across borders.

Our AI Charter keeps that wider ambition open. The current development focus is Evidence-Gated Agents: a bounded attempt to make consequential AI decisions answerable to evidence and authority. Alongside that work, Geneva 2027 remains a possible place to contribute, learn and build connections with the wider public-AI movement. A useful contribution would need a clear purpose, willing collaborators and a scope we can actually support.

A movement with many contributions

There is already a public-AI movement to learn from and contribute to. The Public AI Network brings together a coalition around public access, accountability and lasting public goods. Current AI describes its role as coordinating and supporting public-interest AI across countries, institutions and communities. On the technical side, Apertus, developed through the Swiss AI Initiative by EPFL, ETH Zurich and CSCS, makes model weights, training data, code and methods available for inspection and reuse.

These efforts work at different levels. Their value includes the connections between them: research can become usable infrastructure; public institutions can articulate needs; communities can test whether the result serves them. Our AI Charter is a separate initiative. These references identify related work, without implying affiliation, endorsement or an agreed collaboration.

A constructive contribution starts with understanding what others have built and where a specific question remains unresolved. An evaluation method, a documented failure case, a multilingual comparison or a governance requirement may be useful across several projects. Each can be small enough to examine and improve together.

Public capacity people can use

The Charter's five connected pillars describe the capabilities we want to connect: accountable AI decisions and actions; open, plural models; data and provenance commons; shared assurance and evaluation; and federated public AI infrastructure.

For a user, those categories should translate into practical abilities. To understand where information came from. To choose between models and providers. To know who is responsible for a decision. To have an error examined and corrected. To keep a service working when one component becomes unavailable.

The people affected by AI need a part in shaping these arrangements, alongside those who build, operate and evaluate systems. Shared infrastructure should make room for different languages, circumstances and institutions. Public scrutiny also needs appropriate access: personal or sensitive information may require protected independent review rather than open publication.

For Switzerland, a useful role is to connect scientific capability, public institutions and international cooperation. That role gains value when other societies can adapt the work and contribute on their own terms. Skills, maintainable infrastructure and the ability to change providers are part of the economic benefit: they help people and organisations retain the capacity to act.

A focused contribution: evidence before reliance

Consider an AI-generated recommendation to choose a supplier because a service is supposedly guaranteed. A positive product description would not establish that guarantee. The person relying on the recommendation needs to see the supporting evidence, unresolved qualifications and who has authority to proceed.

Evidence-Gated Agents explores how such checks could be made part of AI-assisted work. The selected prototype would have a normal AI agent propose a decision, retrieve material through an evidence-service interface, analyse its support and limits within EGA using planned FactHarbor reuse, and use controls outside the acting model to release or stop it. The proposed outcome includes an inspectable record of the check. This first integration concerns decision release; it does not execute the resulting action.

The foundations have different levels of maturity. FactHarbor is an Alpha evidence-analysis system. The separate Charter Runtime is an unfinished proof of concept exercised through simulated scenarios and a local test action. The selected EGA integration is not yet implemented. Its usefulness, reliability and limits still need to be demonstrated.

This is one contribution to the wider agenda. The Charter Commitments and evaluation protocol also remain open to criticism and adaptation. Technical checks need accountable people and institutions around them: permission, evidence, independent review and remedy remain distinct responsibilities.

An open path to Geneva

The Geneva AI Summit is scheduled for 21–22 June 2027, with the theme “Bridging Innovation and Trust.” Its stated priorities connect AI's contribution to prosperity and progress with trustworthy, responsible and safe use. That creates a relevant setting for work on public capacity and accountable AI.

There are concrete ways to explore a contribution. The official Road to Geneva programme accepts proposals for independent pre-events. The summit website says approval brings a listing and access to the programme logo, but no funding or organisational support. As checked on 2026-10-04, it also announces a Call for Issues for autumn 2026 and a side-event call for early 2027; those announcements should not be mistaken for open submission windows. Participation details should be checked at the official source before acting.

For Our AI Charter, a contribution could take the form of a focused workshop around a documented case, a comparison of evidence-and-correction methods, or a short public note developed with others. If a demonstrator is ready and independently examined, it could inform that discussion. If the work instead exposes an unresolved problem, a clear account of that limit can still be valuable.

The choice should follow the work and the people willing to carry it. A summit appearance, event, partnership or delivery date has not been secured. Geneva remains an opportunity to explore while development proceeds at a scope that can be justified and supported.

Where cooperation can start

If you are building public AI, researching its effects, providing a public service or representing people affected by its use, a useful starting point is a concrete case: a decision whose basis is difficult to inspect, a correction that does not reach the right people, or a service that depends too heavily on one provider.

We welcome proposals for a bounded contribution around such a case, including contributions to the Geneva process. Technical, institutional and community perspectives can help determine what is worth testing, what already exists, and who should be involved. Public discussion can begin through the repository; proposals or introductions can go to info@factharbor.ch.

The aim is public capacity that people can use, inspect and help govern. Work that earns that trust will remain useful beyond any summit.