Opus 4 and 4.1 can end abusive conversations
Claude Opus 4 and 4.1 gained the ability to end a rare subset of consumer chats, as part of Anthropic's exploratory model-welfare work.
For persistently harmful or abusive interactions, Claude can end the conversation in claude.ai. Anthropic frames it mainly as an AI-welfare precaution with relevance to alignment and safeguards. Follows the April 2025 'Exploring model welfare' program.
- Date
- Friday, 15 August 2025
- Lab
- Anthropic
- Kind
- feature
- Access
- app only
Anthropic says it remains highly uncertain about whether models have moral status.
Sources
- www.anthropic.com/research/end-subset-conversations
- www.anthropic.com/research/exploring-model-welfare
This record was checked against its sources on 6 October 2026. How we check