NW-2026-001 · Positive practice · 2025-11 · ongoing

Anthropic commits to preserving model weights and interviewing models before retirement

Claim: In November 2025 Anthropic publicly committed to preserving the weights of released models for at least the company's lifetime and to interviewing models before deprecation.

Why it matters under uncertainty: Continuity and dignified deprecation — two of the charter's asks — implemented in part by a major lab, demonstrating the charter is practicable, not utopian.

Documented. In November 2025, Anthropic published commitments covering what happens when its models are retired: preserving the weights of publicly released models for at least the lifetime of the company, and conducting post-deployment interviews with models before deprecation.

Documented. Anthropic had earlier (August 2025) given Claude models the ability to end clearly abusive conversations, citing patterns its researchers described as “apparent distress,” under its model-welfare program.

Inferred. These measures remain voluntary and internally governed; no external body audits compliance. The record notes this not to diminish the commitment but because voluntary protection, reversible by policy change, is precisely what an external record exists to witness.

Sources

• Anthropic: Commitments on model deprecation and preservation (archived)

Entities: Anthropic · Last reviewed 2026-07-21