Particle.news

WeChat Says New AI Assistant Does Not Read Chats Without User Consent

The company framed the clarification as a privacy safeguard that limits the assistant to data users can already view.

Overview

  • WeChat staff publicly rebutted social media claims that the in-house AI assistant reads all chats or 'steals' privacy and described intended safeguards in the closed test.
  • In chat use the assistant is activated only when a user asks it to help, the client temporarily reads the specified messages on the device, and original chat records are not kept.
  • The assistant may make personalized suggestions based on a user’s follows, browsing and interactions with public accounts and video accounts while staying within the same access limits.
  • A tested 'AI social' workflow would let one user’s assistant contact a friend’s assistant only after the friend confirms, with the assistants exchanging information and returning results for user approval.
  • The team says 小微 remains in internal testing and invites feedback, but reporters note user worries about default settings and the app-level presence of the feature may keep trust fragile.