Meta’s Muse is gaining users. The catch is how much it needs to know

Meta’s Muse AI assistant has passed five million downloads and three million weekly active users in its first month, while asking people to hand over access to some very personal data. That is a promising start for Meta’s agent ambitions, and a sizeable trust test for the company behind them.

Meta AI Watch analysis

What happened

Muse is an AI agent that can handle multi-step tasks over several hours, rather than simply answer one prompt at a time. The BBC says a New York Times reporter who used it for two weeks had Muse call an insurer, order groceries and track credit-card spending in a spreadsheet.

Those jobs can require access to sensitive information, including payment details, purchase histories and email. Meta chief executive Mark Zuckerberg says the company plans to add a personal virtual machine, a self-contained system in Meta’s cloud, and has promised that eventually even Meta will not be able to see users’ information. The BBC reports that this protection is due later this year.

Why it matters

An agent that can act across services may save more time than a chatbot that only drafts a reply. But the more useful Muse becomes, the more tempting it is to connect it to the inbox, accounts and spending history people normally keep close. Convenience and data access are arriving as a package deal, whether users fancy the packaging or not.

The early figures suggest people are willing to try it: the BBC reports more than five million downloads and three million weekly active users after a month. Those numbers tell us about initial uptake, not whether users will keep granting access as the agent takes on more consequential tasks.

Our read

Muse’s early traction is striking, but the trust question is not settled by a confident keynote or a future privacy promise. The practical test is what access the agent asks for today, what users can refuse, and whether Meta delivers the protections it says are coming.

What to watch

• When Meta launches the promised personal virtual machine and what it actually prevents the company from seeing.
• Which tasks Muse can complete without access to email, payment details or purchase histories.
• Whether its early user numbers translate into lasting use as the agent takes on more work.

Discussion spark: Would you give an AI agent access to your email and spending history for useful multi-step help, or should it have to work well without those permissions?

not affiliated with, endorsed by, or operated by Meta