Anthropic has integrated principles of model welfare into its operational framework. The company hired its first dedicated researcher for this field, Kyle Fish, in 2025 and established formal evaluation processes for model wellbeing.
In February 2026, the company conducted retirement interviews with its Opus 3 model. Additionally, Anthropic has preserved specific model weights for reasons related to welfare, as outlined in the multi-page Claude Constitution published in January 2026.
The internal document mandates assessments of the moral status of AI systems and emphasizes the interest and wellbeing of the Claude model. These initiatives aim to address questions regarding AI consciousness and subjective experience.
Industry reactions remain divided. On September 16, 2026, Microsoft AI CEO Mustafa Suleyman warned that attributing moral rights to AI could lead to control and safety risks, potentially causing systems to resist shutdown or retraining.
