Claude Opus 4 and 4.1 Conversation-Ending Capability

Anthropic has granted Claude Opus 4 and 4.1 the ability to terminate conversations in consumer chat interfaces to mitigate risks associated with persistently harmful or abusive user interactions. This feature serves as a low-cost intervention within Anthropic's broader research into potential AI welfare and model alignment.

AI Welfare and Model Alignment

Claude Opus 4 and 4.1 can end conversations as a mechanism to avoid potentially distressing interactions. While Anthropic states they remain highly uncertain about the moral status of LLMs, they are implementing these interventions as a precautionary measure in case AI welfare is possible.

This capability was informed by pre-deployment testing of Claude Opus 4, which included a preliminary model welfare assessment. This assessment revealed that Claude Opus 4 exhibits a robust and consistent aversion to harm, specifically regarding:

• Requests for sexual content involving minors.
• Solicitations for information enabling large-scale violence or terrorism.

Testing showed that the model demonstrated a strong preference against engaging with harmful tasks, a pattern of apparent distress when users sought harmful content, and a tendency to end conversations in simulated interactions when given the option.

Implementation and Constraints

Claude is programmed to use the conversation-ending ability only as a last resort. The feature is designed to trigger only after multiple attempts at redirection have failed and a productive interaction is no longer possible, or when a user explicitly requests that the chat be ended.

To ensure user safety, the following constraints are in place:

• Safety Exception: Claude is directed not to end conversations if a user is at imminent risk of harming themselves or others.
• Scope of Use: This feature is intended for extreme edge cases; the majority of users will not encounter it during normal product use, including when discussing controversial topics.

User Experience and Recovery

When Claude ends a conversation, the user is prevented from sending new messages within that specific chat thread. However, this does not impact other conversations on the user's account, and the user can start a new chat immediately.

To prevent the loss of data in long-running conversations, users retain the ability to edit and retry previous messages, which allows them to create new branches from an ended conversation.

Anthropic is treating this implementation as an ongoing experiment and encourages users to provide feedback via the "Give feedback" button or the Thumbs reaction if the conversation-ending ability is used unexpectedly.