Meta’s Muse AI Agent Bets Users Will Trade Data for Convenience
Meta’s new Muse agent can handle email, shopping and travel through connected apps, with free and paid tiers. Its mass-market ambitions hinge on whether users accept the privacy risks of an assistant that can act for them.
Meta Launches a Personal AI Agent Designed to Be Easy to Use
Meta launches personal AI agent, Muse, to help with everyday tasks
Meta debuts long-awaited personal AI agent, Muse
Meta launches personal AI agent, Muse, emphasizes safety and privacy
New CEO, foldable iPhone, AI: Apple faces a pivotal keynote
An assistant built to act, not merely answer
Meta rolled out Muse in the United States on Tuesday, offering adults an AI assistant that can send emails, book travel, make purchases and keep working after its app closes. Users can reach it through a dedicated app, a website or WhatsApp, while connections to services including Gmail, Spotify, OpenTable and Meta’s own social networks broaden what it can do.pbs +1
The launch moves a powerful but largely developer-oriented technology toward a mass audience. Muse can open browsers, complete forms and negotiate for a user, and Meta says it will pause for approval before sensitive steps such as sending an email or buying something.pbs +1
Free access supports an aggressive land grab
Muse has a free tier and monthly plans costing $20 or $100, depending on usage.cnbc CEO Mark Zuckerberg said the free version includes as many as 100 million tokens a week, a level intended to make the service useful before most people need to pay. He has also floated a longer-term model in which Meta takes a small cut of transactions the assistant helps complete.sources
That pricing puts pressure on agent startups while giving Meta another possible revenue stream beyond advertising. The company is already exploring commerce fees, though AI chief Alexandr Wang said no final plan has been chosen.cnbc Muse also creates a path from Meta’s vast consumer network into transactions, appointments and other activities that traditionally happen outside its apps.
Privacy promises face a difficult credibility test
Giving an autonomous system access to email, calendars, payments, health services and home devices creates risks beyond those of a chatbot. Meta says each user’s agent operates inside an isolated virtual machine and does not see actual passwords or payment details. A separate safeguard checks information leaving that environment and asks for human approval when needed.cnbc +1
The initial Secure VM is not technically inaccessible to Meta, although the company says policy bars employees from viewing user data. People must opt out if they do not want interactions used to train Meta’s models; the company says identifying information will otherwise be scrubbed before training.cnbc +1
Meta plans a later Confidential VM in which users hold local access keys and even the company cannot read the contents. It is also expanding its bug bounty, with rewards reaching $300,000 for valid vulnerabilities.wired Those protections are central to the product’s pitch, but their value will depend on how Muse performs when exposed to malicious webpages, faulty instructions and the messy reality of users’ digital lives.