Muse is Meta’s Answer to OpenClaw, And it Comes with Trust Issues
Trending Topics Jakob Steinschaden ● Covered by 4 sources
Meta launched Muse, a personal AI agent that can do tasks, not just chat. It starts with email, travel and purchases — and Meta says it still has trust issues.
Based on reporting by Trending Topics, Jakob Steinschaden — 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 launched its biggest consumer AI push yet: Muse, a personal agent that is meant to act, not just answer. It’s rolling out first in the US, on the web at muse.ai, in iOS and Android apps, and inside WhatsApp chats. Meta says its AI glasses are next. Under the hood sits Muse Spark 1.3, the model Meta says recently put it back on top of the benchmarks.
The pitch is broad. Muse is supposed to handle everyday chores and longer projects alike: sending emails, booking trips, renegotiating bills and tariffs, filling out forms, building plans, turning Instagram recipe reels into grocery lists, sending invitations and finishing purchases. It can keep working after the app is closed, then pick back up when something changes or when approval is needed. Payments go through Link by Stripe, and Meta says Muse is the first agent covered by Link’s purchase protections. Shop Pay and 1Password support are planned.
The useful part, of course, is also the risky part. Muse gets access to the services people already use, through a set of connectors that Meta says are added one by one and on purpose. That includes email and calendar, WhatsApp, payments, health and fitness, smart home, dining, shopping, music, events, plus Meta’s own apps like Instagram and Facebook. If there’s a public API, Muse can connect with credentials the user provides. If there isn’t, it uses the browser. For email, users can let it read only, or read and send; permissions can change or be revoked at any time.
Meta is trying to build a fortress around all of this. Each user gets a dedicated Muse Secure VM in Meta’s cloud, where the agent runs and where the data and credentials live. A separate agent called Sentinel approves connector actions and outbound traffic. For sensitive steps like sending an email or making a purchase, approval happens in a dedicated dialog, not in the main chat. The agent never sees passwords or payment details, and Meta says conversations and VM data are not supposed to feed its ad systems.
But there’s a catch, and Meta says it plainly: at launch, the company can still access data when needed to operate, secure or support the service. The stronger Confidential VM, where only the user holds the key, is due before the end of the year. Until then, this is a trust exercise with a very familiar company and a very unfamiliar level of access.
My take — AI-written commentary, not fact-checked reporting
Meta is asking for the kind of trust most companies spend years failing to earn, then act surprised when nobody bites. The agent may be useful, but “we can look inside if needed” is not the selling point it thinks it is. Open AI models are optional; operator access is the real privacy tax, and Europe should keep that in view.
Read more about this at: Trending Topics