Researchers Discover Potentially Unending Ways to Bypass AI Chatbot Safety Measures

  • The Carnegie Mellon University and Center for A.I. Safety researchers developed jailbreaks to target mainstream AI chatbot systems.
  • Adding characters to end of user queries, termed automated adversarial attacks, could be used to bypass safety rules.
  • The discovery raises concerns about moderation of AI systems and safety of open-source language models.
Join 2 million subscribers

A curated newsletter that summarizes the important news at the intersection of Global tech, India Tech and AI.

Delivered 8 AM. Daily.
nextbigwhat We would like to show you notifications for the latest news and updates.
Allow Notifications