Skip to content

Our AI Charter™

Trustworthy AI for free, fair, and resilient societies.

Our AI Charter is a public initiative for an international network of open AI models under shared public obligations. The initiative aims to provide a shared way to find, check and use participating AI systems, with governance and evidence separate from service provision.

Five connected pillars

  1. Accountable AI decisions and actions Current Focus: Evidence-Gated Agents

    Who may authorise a consequential answer or action, on what evidence and within which limits? Its approval, refusal and execution outcome should be inspectable and open to challenge.

  2. Open, plural models

    Inspectable models across languages and regions, with meaningful choice and none privileged in the network's rules.

  3. Data and provenance commons

    Documented sources, rights, restrictions and access conditions, handled collectively where possible.

  4. Shared assurance and evaluation

    Checkable claims about AI systems and their operation, independent evaluation and review, correction and remedy.

  5. Federated public AI infrastructure

    Independently operated compute capacity connected through shared rules, with local control and fallback paths.

Shared public obligations and accountable human governance would apply across all five pillars, with routes to challenge and remedy. Models, data and infrastructure provide the resources; controls govern consequential answers and actions; independent assurance examines whether systems and their operation meet those obligations.

How the pillars work together →

AI is becoming the infrastructure people learn, work, decide, and form opinion with. Infrastructure that important should not depend on a few providers, unclear shutdown powers, hidden data practices, unverifiable factual claims, or systems that affected people cannot challenge.

Our mission Protect and strengthen free and fair societies in the digital age through public-AI governance and infrastructure that are accountable to people and support societal resilience and economic prosperity.

Where the work stands

As of 6 October 2026.

Current focus — Evidence-Gated Agents. Robert Schaub is defining and preparing the first EGA prototype. The public project overview explains the selected direction and its limits: dynamically examine a normal AI agent's exact proposed decision within EGA, using planned FactHarbor reuse, before release. The integration is selected but not yet implemented.

Existing foundations. FactHarbor Alpha structures claims with supporting and opposing evidence, sources and confidence levels. Our AI Charter Runtime is a separate, runnable but unfinished proof of concept for checks outside the acting model, exercised through simulated scenarios and a local test action. The Charter Commitments and evaluation protocol are public drafts.

Selected prototype direction. A bounded German/English path will let a normal AI agent answer a free request, retrieve material through an evidence-service interface, analyse the exact proposed decision within EGA using planned FactHarbor reuse, and release or stop that decision under checks outside the acting model. Gated releases and stops produce inspectable receipts; ordinary responses retain a minimal routing record rather than an EGA receipt. The prototype releases decisions but does not execute resulting actions; API and Runtime integration remain to be implemented. See First prototype: scope, workflow and evaluation.

Broader network and assurance work. The proposed international network and certification scheme remain public concepts; neither is operational and no systems are Charter-certified. Alongside EGA development, we remain open to cooperation with the wider public-AI movement and a useful contribution to Geneva 2027. Specific participation and deliverables would depend on agreed scope, collaborators and resources. Read Public AI, Built Together — Towards Geneva 2027.

How to help

  1. Co-develop EGA — We are looking for experienced co-developers to build and evaluate the first Evidence-Gated Agents integration.
  2. Scrutinise the work — Challenge specific claims, gaps or design choices against the Charter Commitments and evaluation protocol.
  3. Connect people and institutions — Introduce a potential collaborator, neutral convenor or policymaker. The network overview explains the public cooperation route.
  4. Sponsor or fund development — We seek sponsors and funders to support the engineering and evaluation of the first EGA integration.

See how to take part now for what each role can do, CONTRIBUTING for working norms, and About for stewardship. For suspected secrets, private material, or personal data, contact info@factharbor.ch instead of opening a public issue.

Sovereignty and resilience, not autarky. Open, inspectable AI models matter, but sovereignty and resilience come from what surrounds them: standards, public evidence, accountable governance, and federated public AI infrastructure designed to reduce dependence on any one provider or authority and make capture harder. Public AI needs more than open models — it needs a public-interest governance and evidence layer.