Meta Launches AI Agent Muse to Access Apps and Send Payments

Meta launched its Muse AI agent on Tuesday, giving it the ability to send emails, book travel, and make purchases on a user's behalf. The launch came despite internal reports of reliability failures and at least one case of the agent exposing private photos without user permission.

• $18 billion multistate settlement Meta agreed to weeks before the launch involved social media consumer harms.
• 2 paid tiers are available: Power at $20/month and Maximum at $100/month.
• Muse runs on a dedicated virtual machine kept separate from Meta's ads systems, the company says.
• Meta CTO Andrew Bosworth reported being repeatedly logged out during internal testing.

Why it matters: Muse represents a shift from AI that answers questions to AI that acts — connecting to real accounts holding financial, health, and personal data. Errors carry consequences beyond a bad chat response.

• Internal testers reported the agent stopping mid-task, silently ignoring errors, and uploading iCloud photos without permission after a narrow prompt about birthday party toys.
• Meta VP Vishal Shah said the company delayed an April release to improve safety and that the product now meets minimum thresholds — while acknowledging mistakes remain possible.

How 18 sources split on this story

Where they split: Coverage agrees on the facts of the launch but disagrees on how much weight to give Meta's privacy history versus its stated technical safeguards.

What’s next: Meta says Muse will expand to its AI glasses soon, without specifying a date.

• Security experts have not yet independently reviewed Meta's technical privacy claims.
• Will independent security audits confirm or undercut Meta's claims about data isolation and ad-system separation?
• How will regulators, including the FTC, respond to a product that grants Meta access to users' payments, health apps, and email?