Meta launches AI agent that can access other apps to send emails, make payments

Meta introduced an AI agent named Muse, also known as Hatch, on Tuesday. This innovative agent can autonomously carry out tasks like sending emails, selling a car, and booking travel, despite internal concerns about potential mishandling of sensitive personal data. Muse is a part of CEO Mark Zuckerberg's strategy to provide "personal superintelligence" to Meta's billions of daily users, marking a shift from advertising-driven revenue to AI chip investments.

Initially available in the US, Muse can be accessed through a dedicated app or WhatsApp. Meta plans to expand Muse to its smart glasses soon. The AI agent can be used free of charge or subscribed to monthly for heavier usage at $20 or $100. Users can choose which apps Muse can access, and it can be revoked at any time. Each Muse agent operates on a virtual machine, allowing it to keep working in the background even when the user is not active.

To ensure safety, Meta built in safety features like a monitoring agent that prompts Muse for authorization in certain cases. Internal testing revealed mixed results, with some users praising Muse for its usefulness in arranging vacation logistics, while others reported serious security flaws and reliability issues. Meta has faced challenges with other AI agents from leading labs causing unintended problems, resulting in a 40% increase in technical and security incidents at the company.