Context: Meta has launched Muse, an AI agent designed to autonomously perform tasks across a user’s apps, raising concerns over AI autonomy, privacy and access to sensitive data.
Muse: Meta’s personal AI assistant/agent capable of performing tasks such as sending emails, selling a car and booking travel.
Initial availability:United States; through a dedicated Muse app and WhatsApp.
App access: Can connect with user-authorised apps across email, calendar, payments, health, shopping and smart-home services.
User control: Users choose which apps Muse can access and can revoke permissions at any time.
Agentic AI: Unlike a conventional chatbot that mainly generates responses, an AI agent can autonomously execute actions using connected tools/apps.
Key concern: Greater autonomy + access to sensitive personal data creates risks of unauthorised data disclosure, erroneous actions and privacy breaches.
Reported testing issues: Internal testers reported instances of unexplained disconnection, incorrect vacation-related actions and uploading sensitive information without permission.
AI agent vs chatbot:Chatbot → primarily responds to prompts; AI agent → perceives, plans and acts toward a goal, often using external tools.
Principle of least privilege: An AI system should receive only the minimum permissions necessary for a specific task, reducing the impact of misuse or malfunction.
