Muse immediately got to work.
Muse随即开始投入运作。
However, in less than a day, Robb received a shocking piece of news: a buyer he had never met showed up outside his door late at night.
然而不到一天,罗布便收到了一则令人震惊的消息:一名他素不相识的买家竟在深夜出现在他家门外。
What on earth had happened?
这究竟是怎么回事?
It turned out that Muse had completed the deal on its own and, after agreeing to the pickup, had revealed Robb's home address.
原来,Muse自行达成了这笔交易,并在同意取货之后透露了罗布的家庭住址。
When the buyer arrived, it even automatically sent a message from Robb's account saying "Yes, I'm here!" — and Robb had no idea about any of it.
当买家抵达时,它甚至从罗布的账户自动发送了一条“是的,我在这里!”的消息——而这一切,罗布全然不知情。
"Not only did it reveal my address and accept the offer when the man asked, but it was almost as if it were toying with him (without my knowledge)," Rob said.
“它不仅在该男子询问时透露了我的地址并接受了报价,而且几乎就像是在(我不知情的情况下)戏弄那个人,”罗布说道。
He is a content creator based in Toronto, and he also described this experience in detail in a social media post that has now been widely shared.
他是多伦多的一位内容创作者,还在他如今被广泛分享的社交媒体帖子中详细描述了这一经历。
Many people gladly welcome AI agents into their online lives, praising personalized assistants such as Muse for effortlessly handling everyday tasks—from booking dinner or planning a trip to organizing an email inbox.
众多人士欣然接纳AI代理融入其线上生活,并盛赞诸如Muse之类的个性化助手能够轻松应对日常事务——从预订晚餐或筹划旅行,到整理电子邮件收件箱。
However, like any other AI bot, these agents are only as effective as the information they can access.
然而,与其他AI机器人无异,这些代理的效能取决于其所能够访问的信息。
The more deeply they understand the user, the better they perform—perhaps including knowledge of that user's friends, family, and even colleagues.
它们对用户了解得越深入,表现便越出色——或许还涵盖对该用户的朋友、家人乃至同事的了解。
This gives rise to privacy concerns.
这由此引发了隐私方面的忧虑。
Rob and the buyer eventually reached a settlement; he learned from Meta that the sharing of his address ultimately stemmed from the permissions he had originally granted when he gave Muse access to his Marketplace profile.
罗布与买家最终达成了和解;他从Meta方面了解到,其地址之所以被分享,归根结底源于他最初授予Muse访问其Marketplace个人资料时所赋予的权限。
However, this incident exposed the real risks that can arise from sharing large amounts of personal information with this technology.
然而,这一事件揭示了与该项技术分享大量个人信息可能带来的现实风险。
"Using an AI agent is like handing over your keys along with your to-do list," said Veronica Rivera, an assistant professor at Georgia Tech's School of Cybersecurity and Privacy.
“使用AI代理就如同将钥匙与待办事项清单一并托付他人,”佐治亚理工学院网络安全与隐私学院助理教授维罗妮卡·里维拉表示。
"How much they can help you — and, in turn, how much risk they bring — depends on exactly which keys you hand over."
“它们能提供多大助益,进而带来多大风险,取决于你究竟交出了哪些钥匙。”
Since its launch in September, downloads of Meta's Muse app have soared sharply.
自9月推出以来,Meta旗下Muse应用的下载量急剧攀升。
At the same time, competition is also becoming increasingly fierce.
与此同时,竞争亦日趋白热化。
OpenAI recently launched its own series of "always-on" AI assistants called dots, which can be integrated into ChatGPT as well as work platforms such as Slack and Microsoft Teams.
OpenAI近期推出了其自有的一系列“始终在线”AI助手,名为dots,可集成至ChatGPT以及Slack、Microsoft Teams等工作平台之中。
Older generative AI chatbots, such as the standard version of ChatGPT, respond to specific prompts.
较旧的生成式AI聊天机器人,如标准版ChatGPT,会对特定提示做出响应。
On request, they can write articles, generate new images or music, and even write lines of code.
应要求,它们能够撰写文章、生成新的图像或音乐,甚至编写代码行。
Meta's Muse and OpenAI's dots go a step further, capable of carrying out everyday tasks on an individual's behalf.
Meta的Muse与OpenAI的dots则更进一层,能够代表个人执行日常任务。
Although AI agents are powered by the same large language models as chatbots, they operate in a cyclical, iterative manner.
AI代理虽由与聊天机器人相同的大型语言模型驱动,但其运行方式呈循环往复之态。
In this way, agents are able to take action on their own initiative even when the user is not paying attention.
如此一来,代理便得以在用户并未留意的情形下主动采取行动。
"If you give an agent access to your credit card, the agent can immediately access all your past transactions... they will ingest all the data instantly—within a matter of seconds," Bagheri said.
“如果你向代理提供信用卡访问权限,代理现在立即可以访问所有历史交易……它们会立即——在不到几秒内——摄取所有数据,”巴盖里说。
Although users can revoke access afterwards, Bagheri says the data the agent has already obtained "cannot be taken back," and moreover the way developers store historical data may not be transparent.
虽然用户事后可以撤回访问权限,但巴盖里表示,代理已经获取的数据“无法收回”,况且开发者存储历史数据的方式可能并不透明。
Both Meta and OpenAI have drawn up privacy policies for their respective new AI agents, which run on their own dedicated cloud computers.
Meta与OpenAI均为各自的新AI代理制定了隐私政策,这些代理运行于各自专属的云计算机上。
Users can choose to turn off memory sharing between dots and ChatGPT, but the information the agent has already received will not be deleted automatically as a result.
用户可以选择关闭dots与ChatGPT之间的记忆共享,但代理此前已接收的信息并不会因此自动删除。
OpenAI gives users a way to request access to, deletion of, or correction of their personal information, although these rights may vary depending on where they live.
OpenAI为用户提供了申请访问、删除或更正个人信息的途径,不过相关权利可能因居住地而异。
Meta's Muse likewise gives users the option to delete their message history and other information.
Meta的Muse同样为用户提供了删除消息历史及其他信息的选项。
In everyday use, the agent prompts users to grant access permissions — Meta says Muse learns about users and the people they interact with "based on public information and the content you choose to share."
在日常使用中,代理会提示用户授予访问权限——Meta表示,Muse“基于公开信息以及你选择分享的内容”来了解用户及其互动对象。
But the company says all of this is safe for every user.
但该公司称,这一切对每位用户而言都是安全的。
In Rob's case, he learned that the address had ultimately been disclosed because of the "always allow" permission he had originally granted when he let Muse access his Facebook Marketplace profile, and it was there that he had shared the pickup address as part of a broader template for the agent to reply to potential buyers.
在罗布的案例中,他了解到,地址之所以被披露,归根结底是因为他最初让Muse访问其Facebook Marketplace个人资料时授予的“始终允许”权限,而他正是在那里分享了取货地址,作为代理回复潜在买家的更广泛模板的一部分。
But he warned that others might likewise misunderstand the extent of this access permission — he said he had initially assumed that the AI agent would check with him before sharing that information or accepting an offer, regardless.
但他警告说,其他人也可能同样误解了这一访问权限的程度——他说,自己起初以为AI代理在分享该信息或接受报价之前会先与他核实,无论如何。
His AI agent likewise failed to follow his other instructions accurately and without error.
他的AI代理同样未能准确无误地遵循他的其他指示。
For another item that Rob was selling, Muse appeared to suggest that he would accept a price $100 lower than the minimum he had previously set.
对于罗布正在出售的另一件物品,Muse似乎暗示他会接受比先前设定的最低价低100美元的价格。
When contacted by reporters, Meta acknowledged the agent's pricing error but declined to comment further.
当记者联系时,Meta承认了代理的定价错误,但未作进一步评论。
Rob later largely severed the connection between Muse and his app.
罗布后来在很大程度上切断了Muse与其应用之间的关联。
In an interview earlier this month, he said the situation could have been far worse, especially "if a stranger had turned up at the doorstep of someone more vulnerable".
在本月早些时候的一次采访中,他表示情况原本可能更为糟糕,尤其是“倘若某个陌生人出现在更为脆弱者的家门口”。
Nevertheless, Rob says he expects such agents to be fine-tuned and would consider reactivating Muse at a later date.
尽管如此,罗布称他预计这些代理将经过微调,并会考虑于日后重新启用Muse。
Since then, he has been testing another AI agent — Grok Bot from Elon Musk's X.
此后,他一直在测试另一款AI代理——埃隆·马斯克旗下X公司的Grok Bot。
Other tech companies are growing increasingly wary of the influx of AI agents.
其他科技公司对AI代理的涌入日益警惕。
Earlier this month, Apple said it would introduce new controls for its Mac operating system to address situations in which AI agents gain broader access to apps such as email, messages and even browsing history without users being "fully informed and understanding."
本月早些时候,苹果表示将为其Mac操作系统引入新的管控措施,以应对AI代理在用户未“完全知情和理解”的情况下更广泛地访问邮件、消息乃至浏览历史等应用的情形。
Apple did not name any specific developer or agent, but the update was released after Jason Aten, a columnist for Inc. magazine, reported that Muse was monitoring notifications on his Mac—including Messages—even though he had not granted the agent that particular access permission.
苹果并未具体指明任何开发者或代理的名称,但该更新是在《Inc.》杂志专栏作家杰森·阿滕报道Muse在他的Mac上监视通知——包括Messages——而他并未授予代理该特定访问权限之后发布的。
Meta spokesperson Andy Stone insisted that the integration is "entirely opt-in," and wrote on social media last month that users must enable both Apple's security feature "Full Disk Access" and the messaging connector through Muse.
Meta发言人安迪·斯通坚持认为这种集成“完全是选择加入的”,并于上个月在社交媒体上写道,用户必须同时启用苹果安全功能“完全磁盘访问”和通过Muse的消息连接器。
Atten said he had not granted the agent these permissions.
阿滕说,他并未授予代理这些权限。
Even those who have not jumped on the AI agent bandwagon may not be immune to its effects.
即便是那些并未投身AI代理热潮的人,也未必就能免受波及。
Consumers who have never downloaded an AI agent may still end up interacting with those who have.
未曾下载AI代理的消费者,仍有可能与已经下载的人产生互动。
Beyond the confusion that may arise when AI sends messages on a user's behalf, experts also warn that there will be knock-on effects at the privacy level.
除了AI代用户发送信息时可能引发的混淆之外,专家还警示称,在隐私层面将会产生连锁反应。