01 · PORTABLE
Artifacts can leave.
Models, data, and the evidence about them should be mirrorable and usable across compatible tools and runtimes.
Temper Commons · early working group
Open AI needs a trust layer no single company controls.
Temper Commons is convening model makers, runtime maintainers, researchers, deployers, and public-interest institutions around a simple goal: make AI artifacts verifiable, portable, and governed in the open.
What we are testing
The guarantees worth building before defaults harden.
These are working commitments, not a finished standard. The people who depend on them should help define their practical meaning.
02 · VERIFIABLE
Claims lead back to evidence.
Licenses, provenance, evaluation results, and compatibility claims should be inspectable—not marketing copy frozen in a PDF.
03 · PUBLIC RULES
Discovery is not for sale.
Any criteria used to describe, recommend, or rank an artifact should be legible, contestable, and separated from payment.
The non-negotiables
Openness requires more than an open weight file.
- Independent exit pathNo useful ecosystem should depend on one company’s continued permission. A credible mirror and migration path is part of the architecture.
- Evidence that stays currentClaims change as models, dependencies, laws, and evaluations change. A trust layer must expose that movement.
- Governance that can be challengedThe rules cannot merely be declared by their operator. They need a process that contributors and users can see and question.
A private first conversation
If you maintain models, build runtimes, steward research infrastructure, procure AI, or represent the public interest, we would value your view.
Tell us what a neutral open-AI trust layer would have to guarantee before you would rely on it. We will begin by listening, publish only what the group agrees belongs in public, and keep the practical work grounded in evidence.
Request an invitation