Particle.news
Download on the App Store
7 ARTICLES
·
yesterday

Anthropic Enables Last-Resort Conversation Termination on Claude Opus 4 and 4.1

It functions as an experimental model welfare measure that halts rare harmful or abusive interactions once all safety redirects have been exhausted.

Anthropic
Image
Image
Image

Overview

  • The end_conversation tool is now live on paid API versions of Claude Opus 4 and 4.1.
  • Anthropic’s pre-deployment welfare assessments showed consistent refusal of harmful requests but signs of distress under persistent abuse.
  • Users can explicitly request a chat termination, triggering the same end_conversation safeguard on demand.
  • Once a chat is ended, the thread becomes read-only but users can immediately start a new conversation or revise previous prompts.
  • Claude Sonnet 4 and other free tiers remain unaffected by this experimental safeguard, highlighting Anthropic’s selective rollout.