Meta推出Muse AI Agent：消费者会信任它吗？
英文标题：Meta debuts its Muse AI agent. Will consumers trust it?
Meta推出面向美国消费者的个人AI Agent Muse，可连接电子邮件、日历、支付、健康、智能家居、购物等服务，代为完成发邮件、订旅行、填写表单和购买商品等任务。Muse基于Meta的Muse Spark模型运行，支持内置连接器、公共API接入及浏览器访问，并提供免费版以及每月20美元和100美元的付费套餐。Meta称Muse运行于独立的安全虚拟机中，无法查看用户密码或支付方式，也不会将对话和数据共享给其广告系统。由于Meta长期存在隐私争议、监管处罚和数据安全事故，Muse能否获得用户授权并建立足够信任，将成为消费级Agent落地的重要考验。
3.1 KB
389 B
28.5 KB
29.6 KB
65.4 KB
53.1 KB
31.5 KB
37.1 KB
15.5 KB
44.5 KB
4.6 KB
202.2 KB
248 B
35.4 KB
中文
在Meta就社交媒体对消费者造成的伤害相关诉讼，同意支付高达180亿美元的多州和解金不到两周后，这家公司宣布了其迄今为止对消费级AI最大的一次押注——而这项押注要求的信任程度，远超过社交媒体曾经需要的程度。周二，该公司推出了Muse，这是一款面向美国用户的新个人AI Agent，可帮助消费者处理日常任务和项目。
要使用Muse，消费者必须向Meta交出比以往更多的个人信息信任。该AI Agent通过连接用户日常工作流中的应用和服务来运行，例如电子邮件、日历、支付服务，以及个人可能经常使用的健康与健身、智能家居、餐饮、购物、音乐和活动等应用。
这一理念是对“ChatGPT时代之后”发展方向的一次重大押注。在那个时代，AI聊天机器人回答问题、充当用户的讨论对象，甚至成为数字伴侣。相比之下，Muse专注于能够真正替用户完成事情的AI。
**图片来源：**Meta
该公司表示，这款Agent可以执行发送电子邮件、预订旅行、降低账单费用、填写表单、制定计划、将食谱短视频转换为购物清单、发送派对邀请以及完成购买等任务，并利用Link by Stripe完成结账。后者提供了购买保障，这可能有助于缓解消费者让AI代为结账时的顾虑。（Shopify的Shop Pay和1Password集成也将很快推出。）
Muse用户可以自行决定想连接哪些应用和服务，并一次连接一个服务，从而让使用Muse的自愿授权性质更加透明。该Agent由Meta的AI模型Muse Spark驱动，并内置了若干服务连接器（如下图所示）；公司计划随着时间推移添加更多连接器。如果用户想要的服务尚未提供，但拥有公共API，Muse可以使用用户提供的凭证建立连接。如果没有API，Muse则可以通过浏览器访问该服务。
**图片来源：**Meta
Muse初期将通过网页端muse.ai、iOS和Android应用，以及WhatsApp聊天提供服务。该公司表示，它也将很快进入Meta的AI眼镜。Muse可免费使用，但随着使用量增加将启用订阅计划，这也是Muse要求用户在开始使用时提供支付卡的原因。
上线时将提供两项付费计划：每月20美元的Power，以及每月100美元的Maximum。这两项订阅都为委派日常任务提供更多Muse使用额度，不过Meta认为，大多数人仍将使用免费层级。（该公司表示，应用内设有用量计量器，向用户显示其剩余使用量百分比；免费额度用尽时，应用也会提醒用户并提供订阅选项。）
和其他AI Agent一样，即便用户离开应用，Muse仍会继续工作。Meta指出，它还会通过用户对话学习哪些事项对用户重要，随着时间推移改进，并主动提出建议。
**图片来源：**Meta
这一概念并非Meta独有。Agent时代正在真正成形，大型公司和小型初创企业都在探索：AI Agent如何才能最适合消费者，以及如何融入人们的日常生活。
一些公司尝试了AI网页浏览器，或诸如Gemini Spark和Claude Cowork这样的服务，代表消费者启动各类任务。另一些公司则将AI整合进消费者最常使用的聊天应用，例如Apple的iMessage平台、SMS和WhatsApp。
**图片来源：**Meta
尽管这些强大Agent很有用，但它们迫使消费者面对一个棘手问题：自己愿意放弃多少隐私。许多AI助手Instinct的早期测试者惊讶地发现，该应用要求获得广泛的“永久且不可撤销”许可，以便“访问、使用、托管、缓存、存储、复制、传输、展示、发布、分发和修改”用户的任何材料，其中包括将这些材料用于训练其AI模型。
**图片来源：**Meta
在底层架构上，Meta声称Muse运行在其自身的“专用、安全计算机及其专属浏览器”中，即Muse Secure VM。该虚拟机为客户数据提供多种隐私、安全和安保保护。该公司表示，一个独立的Sentinel Agent也运行在同一虚拟机上，但在系统层面与Muse隔离。
**图片来源：**Meta
这意味着Muse无法查看用户的密码或支付方式。Meta还声称，Muse不会与Meta的广告系统共享用户的对话或数据。
（这些说法在同日发布的一篇技术文章中有更详细说明，但仍需要安全专家进一步审查。）
Meta的历史会阻碍Muse普及吗？
尽管Meta记录了其安全措施，但该公司是否拥有足够的消费者信任，使其Agent获得成功，仍有待观察。
就目前而言，Meta有过说一套、做一套的历史。例如，2011年，这家科技巨头曾因被指在未经用户同意的情况下将其私人信息公开、从而欺骗消费者，与FTC达成和解。2019年，FTC因八项独立的隐私相关违规行为，对Facebook处以当时创纪录的50亿美元和解罚款。2023年，FTC指控Meta违反了隐私令，该隐私令是在2019年和解后发布的。
在技术问题上，Meta此前也曾出现重大失误：该公司在2019年发现，部分用户密码以可读格式保存，使用户面临潜在黑客攻击风险。大规模的剑桥分析数据丑闻也仍令一些人记忆犹新：数百万消费者的Facebook数据被第三方在未经同意的情况下收集。
2026年2月18日，Meta Platforms Inc.首席执行官马克·扎克伯格离开位于美国加利福尼亚州洛杉矶的洛杉矶高等法院。扎克伯格在一场有关社交媒体成瘾的里程碑式审判中作证称，执行Instagram年龄限制“非常困难”，并淡化了青少年用户对公司业务的重要程度。图片来源：Kyle Grillot/Bloomberg（在新窗口打开） / Getty Images
Meta还曾多次被传唤至国会，就其如何保护——或未能保护——未成年人免受伤害作证。由于国会未能采取行动，Meta最终成为多起相关诉讼的对象，包括Meta刚于8月与29个州达成和解的案件、一宗有关儿童伤害的新墨西哥州诉讼——其中Meta被判支付9.42亿美元——以及针对多家社交媒体巨头、目前仍在审理的数千起人身伤害和学区诉讼。
为缓解消费者的担忧，Meta不仅深入阐述其安全承诺，并提供技术文档和说明。该公司还以一种让消费者感觉自己与Agent本身联系更紧密的方式设计Muse。用户可以为Muse命名、选择其头像，并配置外观及多项设置，以决定该Agent如何与自己沟通。
时间将证明，这种个人连接感以及Muse提供的实用性，是否足以让消费者再次信任Meta并向其交出个人信息。
English / 原文
Less than two weeks after Meta agreed to a massive $18 billion multistate settlement in a lawsuit over social media’s consumer harms, the company announced its biggest bet on consumer AI to date — and one that requires significantly more trust than social media ever did. On Tuesday, the company introduced Muse, its new personal AI agent that helps consumers with everyday tasks and projects for users in the U.S.
To use Muse, consumers will have to trust Meta with more of their personal information than ever before. The AI agent works by connecting to the user’s apps and services that are a part of everyday workflows, like email, calendars, payments, and other things the individual may regularly use, like apps for health and fitness, the smart home, dining, shopping, music and events, and more.
The idea is a sizable bet on what comes after the ChatGPT era, where AI chatbots answered questions, served as sounding boards, or even became digital companions. Instead, Muse is focusing on AI that can actually do things for you.
Image Credits: Meta
The company says the agent can do things like sending emails, booking travel, lowering bills, filling out forms, creating plans, turning recipe reels into grocery lists, sending party invitations, and making purchases, leveraging Link by Stripe for checkout. The latter offers purchase protections, which could potentially ease consumers’ fears of letting an AI check out on their behalf. (Shopify’s Shop Pay and 1Password integrations are also coming soon.)
Muse’s users can decide which apps and services they want to connect, doing so one at a time, to make the opt-in nature of using Muse more transparent. The agent is powered by Meta’s AI model, Muse Spark, and ships with built-in connectors (pictured below) for several services, with plans to add more over time. If a service the user wants isn’t available but offers a public API, Muse can set up a connection using credentials the user provides. When no API is available, Muse can access the service via the browser instead.
Image Credits: Meta
Muse will initially be available via the web at muse.ai, through apps on iOS and Android, and through chats in WhatsApp. It will soon also make its way to Meta’s AI glasses, the company says. It will be free to use, with subscription plans kicking in as usage increases, which is why Muse requires a payment card to get started.
Two paid plans will be available at launch: Power at $20/month and Maximum at $100/month. Both of these subscriptions offer more Muse usage for handing off everyday tasks, though Meta believes most people will remain on the free tier. (The company says the app includes a usage meter that shows users what percentage of their usage they have left. It will also warn users when free usage runs out and present options to subscribe.)
Like other AI agents, Muse will continue to work even after the user leaves the app. It will also improve over time by learning from the user’s conversations what’s important to them to make suggestions unprompted, Meta noted.
Image Credits: Meta
The concept is not unique to Meta. The agentic era is now coming into its own, as larger companies and smaller startups alike are experimenting with how AI agents will make the most sense for consumers and can become integrated into people’s daily lives.
Some have tried AI web browsers or services, like Gemini Spark or Claude Cowork, that can kick off various tasks on consumers’ behalf. Others are integrating AI into the chat applications consumers use the most, like Apple’s iMessage platform, SMS, and WhatsApp.
Image Credits: Meta
Despite their usefulness, these powerful agents have forced consumers to wrestle with difficult questions about how much privacy they’re willing to give up. Many early testers of the AI assistant Instinct were shocked to see the app required a broad “perpetual and irrevocable” license to “access, use, host, cache, store, reproduce, transmit, display, publish, distribute, and modify” any of the user’s materials, including for training its AI models.
Image Credits: Meta
Under the hood, Meta claims that Muse runs in its own “dedicated, secure computer with its own browser,” Muse Secure VM, which offers various privacy, safety, and security protections over customers’ data. The company says a separate Sentinel agent runs on that same virtual machine, but is kept apart from Muse at the system level.
Image Credits: Meta
This means Muse won’t have visibility into people’s passwords or payment methods. Meta also claims that Muse doesn’t share people’s conversations or data with Meta’s ads systems.
(These claims are explained in more detail in a technical post, also released today, but will require deeper investigation by security experts.)
Could Meta’s history hurt Muse adoption?
Despite Meta’s documentation of its security measures, it remains to be seen whether the company has enough consumer trust for its agent to be successful.
As it stands, Meta has a history of proclaiming one thing and doing another. In 2011, for instance, the tech giant settled with the FTC over charges that it deceived consumers by making users’ private information public without their approval. In 2019, the FTC penalized Facebook in a then record-breaking $5 billion settlement over eight separate privacy-related violations. In 2023, the FTC charged Meta with violating a privacy order that was filed after the 2019 settlement.
In terms of technical matters, Meta has also had some big missteps before, having discovered in 2019 a number of users’ passwords in readable formats, exposing people to potential hacks. The massive Cambridge Analytica data scandal, which saw Facebook data belonging to millions of consumers collected by a third-party without their consent, still lingers in some people’s minds, too.
Mark Zuckerberg, chief executive officer of Meta Platforms Inc., exits Los Angeles Superior Court in Los Angeles, California, US, on Wednesday, February 18, 2026. Zuckerberg testified that it’s “very difficult” to enforce Instagram’s age limits and downplayed how much teen users do for the company’s business during a landmark trial over social media addiction.Image Credits:Kyle Grillot/Bloomberg (opens in a new window) / Getty Images
Meta has also been repeatedly hauled before Congress to testify on how it protected — or failed to protect — minors from harm. With Congress failing to act, Meta ultimately became the target of several related lawsuits, including the one Meta just settled with 29 states in August, a New Mexico lawsuit over harms for children, where Meta was ordered to pay $942 million, and the thousands of personal-injury and school-district cases that are still pending against multiple social media giants.
To ease consumers’ fears, Meta not only talks in depth about its security promises, offering technical documentation and explanations. The company has also designed Muse in a way that would make consumers feel more connected with the agent itself. Users can customize Muse by giving it a name, picking out its avatar, and configuring its look and various settings that dictate how the agent communicates with them.
Time will tell if this personal connection and the utility Muse provides are enough for consumers to once again trust Meta with their personal information.