Meta’s Muse AI agent is drawing attention for moving beyond the traditional chatbot model. Launched on September 8, Muse is designed to carry out multi-step tasks on a user’s behalf, including organising emails, filling forms, planning trips, shopping and handling other online activities.
What makes Meta Muse different?
Unlike a chatbot that mainly responds to prompts, Muse is designed to take action. Meta says it can work across services such as email, calendars, payments, dining and shopping. It uses its own browser through a dedicated secure virtual machine and can ask users for confirmation before sensitive actions such as purchases or sending emails.
The agent can also interact with services that do not provide a public API by using their websites through a browser, similar to how a person would navigate them.

Amazon blocked Muse from shopping on Amazon.com after saying the AI agent was accessing its platform without authorisation. Amazon also raised concerns that Muse did not identify itself as an AI agent while browsing and could interact with customer accounts and credentials.
Amazon said third-party agents making purchases on behalf of customers should operate openly and respect individual service providers’ decisions.
Meta, meanwhile, has said Muse does not have visibility into users’ passwords or payment information and that credentials are stored securely for authorised use.

The Amazon dispute came alongside disclosure of a zero-day vulnerability in Muse’s macOS application. Security researcher Patrick Wardle found that locally running applications or terminal commands could manipulate an undocumented Muse setting and potentially redirect its transcription process to an attacker-controlled server. Meta subsequently issued a fix, according to reporting.
The developments highlight a larger challenge for agentic AI: the more authority an AI system receives to act across websites, accounts and devices, the greater the need for strong controls over identity, permissions, privacy and security.