Meta’s New AI Agent Is an Instant Hit—and the Backlash Has Already Begun
Meta wanted to build an AI assistant that would do more than answer questions.
It wanted one that could actually get things done.
That vision arrived this month in the form of Muse, Meta’s new personal AI agent. Instead of simply responding to a prompt, Muse can send emails, book appointments, organize schedules, shop online, handle customer-service interactions and work on tasks after the user has closed the app.
À
À propos de Facebook
+1
The response has been immediate.
Muse quickly climbed to the top of Apple’s U.S. App Store, with Reuters reporting more than 2.5 million downloads roughly two weeks after launch. The enthusiasm has given Meta an important early success in consumer AI.
R
Reuters
But the same capabilities that make Muse exciting are also creating its biggest problem.
An AI that can act for you needs considerably more access than an AI that merely talks to you.
From chatbot to digital agent
The fundamental difference between Muse and a conventional chatbot is action.
A chatbot can tell you which flight might be suitable. An agent can search for the flight, fill out forms and potentially book it.
A chatbot can explain how to cancel a subscription. Muse can attempt to do it.
A chatbot can help create a shopping list. Muse can turn a recipe into a list and then help purchase the ingredients.
Meta says Muse operates through a dedicated virtual machine designed to isolate the agent and the user's information. The company says credentials are stored securely, that Muse does not see passwords or payment information directly, and that users must approve sensitive actions such as sending emails or making purchases.
À
À propos de Facebook
That architecture is central to Meta's pitch.
The company is effectively asking consumers to believe that an AI with significant access to their digital lives can be given enough power to be useful without giving it enough power to become dangerous.
Then Amazon said no
The first major corporate confrontation arrived quickly.
Amazon blocked Muse from shopping on Amazon.com, saying the assistant was an unauthorized AI agent and violated the site's conditions of use. Amazon said Meta had not obtained permission to access the service in this way.
F
Fortune
+1
The dispute exposes a fundamental problem with the agent model.
Today's internet is designed around humans interacting with websites and applications. Agents change that relationship.
Instead of a customer opening Amazon, searching for products, viewing advertisements and completing a transaction, an AI agent can potentially perform much of that process on the customer's behalf.
For Amazon, that isn't merely a technical distinction. The company's business depends heavily on the relationship between consumers, its marketplace and the commercial systems surrounding it.
The dispute therefore raises a much bigger question: Who gets to decide whether an AI agent is allowed to act on a website?
The security problem
The backlash isn't limited to corporate access.
Security researchers have also scrutinized Muse's design.
Ars Technica reported a serious vulnerability in the macOS version that could allow locally running applications or commands to manipulate Muse and potentially take control of an authenticated Muse account. Meta subsequently said it had released a hotfix.
A
Ars Technica
The incident is particularly significant because of what Muse is designed to do.
A security flaw in a conventional app might expose information.
A flaw in an AI agent with access to email, calendars, shopping accounts and other services could potentially allow an attacker to make the agent perform actions on the user's behalf.
That raises the security standard considerably.
The more capable an agent becomes, the more damaging a successful compromise could potentially be.
Privacy becomes harder, too
There is also a basic privacy question that doesn't disappear simply because an agent is technically secure.
For Muse to be genuinely useful, users have to tell it about their lives.
That can include calendars, email, shopping preferences, travel plans and other personal information. Meta says users control which services Muse can access and can disconnect those services or tell the system to forget information. The company also says interactions can be excluded from AI-training use.
À
À propos de Facebook
But consumers still have to decide how much access they are comfortable granting.
That decision is particularly consequential for Meta because the company already has a long history of scrutiny over privacy and data practices.
Muse therefore isn't asking consumers for the same kind of trust as a search engine or chatbot.
It is asking for something closer to delegation.
Meta's bet on convenience
That is also precisely why the product could be so compelling.
Most people don't need another place to ask questions. They need things done.
Appointments have to be scheduled. Bills have to be negotiated. Returns have to be processed. Travel has to be organized. Emails have to be answered.
If an AI can reliably handle those chores, the value proposition becomes much clearer than simply generating another paragraph of text.
Meta is betting that consumers will eventually regard an AI agent the way they now regard cloud storage, mobile payments or digital calendars: as infrastructure that quietly handles part of everyday life.
The company is also extending Muse beyond the phone. At its latest event, Meta introduced additional hardware and said Muse would come to its smart glasses, expanding the possibility of interacting with the agent without opening a conventional app.
L
Los Angeles Times
+1
The human factor
There is an unexpected complication in that expansion.
Reuters reported that Meta has been testing a “human concierge” component for some Muse phone calls, in which human contractors can handle certain calls placed through the service. Meta says the testing is intended to improve safety and privacy and that feedback has been largely positive.
R
Reuters
The development illustrates how difficult fully autonomous agents remain.
When an AI reaches the edge of what it can reliably do, there is a temptation to put a human behind the scenes.
That may make the system more useful.
It also creates a new privacy question: if users believe they are dealing with an AI, when—and how—are humans involved?
The bigger battle
Meta's early success with Muse matters beyond Meta.
Google, OpenAI, Amazon, Apple and other technology companies are developing their own approaches to AI agents. Some are focused on shopping, others on productivity, smartphones or broader personal assistants.
The emerging competition is not simply about which company has the smartest model.
It is about which company can persuade people to let an AI act on their behalf.
That requires three things at once: capable AI, secure infrastructure and user trust.
Muse appears to have demonstrated the first part of that equation. Its rapid adoption suggests consumers are interested in agents that actually perform tasks.
The backlash demonstrates how difficult the other two parts will be.
The real test begins now
The novelty of Muse is likely to fade.
What will matter after that is whether the system can consistently complete tasks without making costly mistakes, whether companies such as Amazon are willing to permit agents to interact with their services, and whether consumers are comfortable granting AI systems access to increasingly intimate parts of their digital lives.
Meta has presented Muse as a step toward a future in which everyone has a personal AI capable of pursuing goals on their behalf.
The early numbers suggest people are willing to try it.
The backlash suggests they—and the companies whose services Muse touches—are not yet ready to give it unlimited freedom.
0 comments:
Post a Comment