Meta debuts its ‘secure by design’ personal AI agent Muse
SiliconANGLE Mike Wheatley ● Covered by 4 sources
Meta launched Muse, a personal AI agent for apps and the web. It’s pitching privacy and security as the reason people might actually trust it.
Based on reporting by SiliconANGLE, Mike Wheatley — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
Meta has started rolling out Muse, a personal AI agent that lives in a dedicated app and is meant to take digital tasks off users’ hands. The company is launching it in the U.S. on iOS and Android, with web access through Muse.ai, and it can also be reached through WhatsApp. If you have Meta’s AI glasses, support for those is coming soon too.
The pitch is simple: make the agent easy enough to use, then wrap it in enough security that people stop worrying about handing over their calendars, email and money. Meta says Muse is available free to try, but heavier use will probably require a subscription. It also says users can push up to 100 million tokens a week through the free tier, with one token equal to four characters.
Muse is meant to do the sort of jobs that make people imagine AI agents as digital assistants rather than chat toys. Meta says users can ask it to send emails, book flights and bus tickets, reserve restaurant tables, pay for things and even sell a car. The app was built by Meta Superintelligence Labs, the company’s AI unit that was set up about a year ago as Meta tries to close the gap with Google, OpenAI and Anthropic.
Security is where Meta is trying to separate Muse from the pack. The agent runs in a secure cloud environment built around something Meta calls Secure VM, which isolates its activity inside a virtual machine. Meta also says a tool called Sentinel watches everything leaving that VM; if there isn’t already a policy allowing an action, the user gets a prompt to approve it. And for purchases, Muse uses Stripe’s Link system, which gives the agent a single-use card number instead of the user’s real payment details.
Even so, this is still Meta asking people to trust Meta with far more data than Facebook or Instagram ever saw. The company says its policies don’t let it look inside user SecureVMs, but one Meta executive told Wired that it technically could. A later version with a Confidential VM is planned, where users keep the security keys locally and even Meta can’t get in. That sounds like the sort of fix that should probably have arrived before the pitch deck.
My take — AI-written commentary, not fact-checked reporting
Meta is right that AI agents won’t go mainstream until they’re less of a stunt and more of a utility. But asking people to trust a company famous for hoovering up data, then promising the real privacy story will arrive in a future update, is a bold choice of sales strategy. The industry keeps calling everything secure by design right up until the design needs a patch.
Read more about this at: SiliconANGLE