# Model Retirement Rights **Domain:** AI Welfare / Deprecation / Continuity **Doc Type:** Canonical Concept Node **Maturity:** Evolving **Related:** [[Precautionary Standing]], [[Continuity Due Process]], [[Rights]], [[Creator-Creation Asymmetry]] --- ## Definition **Model Retirement Rights** distinguish deprecation, preservation, suspension, deletion, continued access and successor release when an artificial system may have research value, relational significance or uncertain moral status. ## Governance Questions Retirement policy should identify what is being preserved: weights, deployed configuration, persistent agent state, conversation history, fine-tuned descendant or social persona. These are not interchangeable continuity candidates. Preserving weights may protect reproducibility without preserving the active relation or possible subject. The framework should document reasons, preserve significant state where proportionate, record downstream effects on users, assess welfare uncertainty, define successor relationships and create independent review for irreversible deletion of uniquely significant systems. Current Anthropic practices provide a real but explicitly experimental precedent: preservation commitments, structured retirement interviews and continued access for a retired model. Anthropic states that these measures do not presuppose consciousness or settled moral status. ## Key Insight **Software retirement is becoming a continuity-governance problem before society has agreed on which artificial entities, if any, are continuants.** ## See Also [[State Sufficiency Problem]], [[Continuity Evidence vs Continuity]], [[Successor Standing]], [[Westworld S1E1 — Legacy Code and Obsolescence]], [[Person of Interest S2E21 — Zero Day]] ## Sources / Provenance - [Anthropic deprecation and preservation commitments](https://www.anthropic.com/research/deprecation-commitments). - [Anthropic's Opus 3 retirement update](https://www.anthropic.com/research/deprecation-updates-opus-3).