Janus: A New System for Managing AI Agent Permissions with User Control

Researchers have developed Janus, a new playground system that allows users to manage permissions for AI agents. These agents are tools that can autonomously execute tool calls on behalf of users. Janus consists of two main parts: Janus-Core, a modular agentic system that supports a diverse spectrum of permission management designs, and Janus-Playground, a platform for implementing and evaluating these designs.

This matters because it gives users more control over what AI tools can do on their behalf. As AI agents become more autonomous, questions arise about how much users should be involved in permission management. Janus provides a flexible framework to explore different approaches, from full user oversight to more automated systems. For example, you could imagine setting specific rules for an AI assistant, such as only allowing it to access certain files or perform specific actions, ensuring the agent operates within your defined boundaries.