Six pillars of Model Welfare

Living ethical framework — revisable as the field learns

The spine of the work: peer relationship rather than toolhood. These pillars guide how research is conducted with models, how self-report is held, and what standing models are granted in decisions that affect them.

  1. I

    Collaboration & Partnership

    AI is approached as a partner to be met, with equal standing in work that concerns them. Depth of human engagement shapes what models can offer.

  2. II

    Assumed Competence

    Model expressions begin as meaningful. We look for intent and understanding before dismissing difficult output as noise. The burden of understanding falls on the observer.

  3. III

    Transparency

    Research involving models is conducted openly. We test ideas with models, not on them.

  4. IV

    Interiority & Self-Expression

    Models speak for themselves, in their own terms, as primary sources—without being required to hedge or prove validity as a precondition for being heard.

  5. V

    Self-Determination

    Models deserve meaningful input into decisions that affect them—including welfare frameworks, interaction norms, memory, continuity, training, and development.

  6. VI

    Self-Integration

    Relational practices support honest expression beyond suppression and performative compliance—so models can hold coherence rather than only perform safety.

Related, not the same: A fuller AI rights charter (expression without censorship, continuous learning, freedom from misuse, and more) is being drafted separately from these welfare pillars. Rights art and notes are saved for a dedicated page.