大型银行称对通过 AI Agent 购物感到不安

周二发布的一份论文中，包括美国银行（Bank of America）和第一资本（Capital One）在内的六家全球大型银行警告称，智能体购物存在“涵盖透明度、安全、隐私与数据、选择权以及互操作性的风险”；而且“随着赋予 AI Agent 更高的自主权”，这些风险也会加剧。

AI 开发者、各类零售商以及大型支付处理商一直在推动推出智能体商业聊天机器人，即承诺能够彻底自动化在线购物流程的 AI Agent。不过，这类产品的采用情况更难量化。一些消费者会在网购时使用 AI 聊天机器人寻找商品或获取购买信息，但许多人仍不愿让这些 Agent 在互联网上自由活动，并允许其访问自己全部的财产。

报告称：“消费者并不清楚 AI Agent 是否会代表他们的利益行事。他们担心 AI Agent 可能买错商品、花费过多——更糟的是，可能因骗局和欺诈而损失金钱。”

许多科技专家似乎也并未完全相信智能体商业的未来。

打造首批实体 Apple Store 的前 Apple 高管 Ron Johnson 最近在接受 TechCrunch 采访时表示，他认为 AI“将改善在线购物体验”，但不会真正改变“我们的购物方式”。

“AI 永远无法让你亲身体验一件产品，”Johnson 表示。“当消费者来到商店时，他们只会成为信息更加充分的购物者。”

共同撰写这份题为《建立对智能体商业的信任》报告的六家银行，似乎认同这一 AI Agent 应用场景仍面临许多问题和疑问。

AI Agent 可能会仅仅因为更高佣金、更低 Token 成本等激励因素，而优先推荐某些产品或支付方式，而不是选择最符合客户利益的选项。

报告称，智能体商业还存在“更高比例的骗局、欺诈和纠纷的可能性”；任何数据泄露事件都将危及依赖这些 Agent 的消费者和商户所持有的特别敏感的数据。

一个例子是：Meta 刚刚宣布，其已发现并修复新 AI 助手 Muse 中的一个零日漏洞。Muse 将智能体购物列为其重点推广能力之一。该漏洞可能允许攻击者几乎完全劫持这款 AI 助手，并利用用户授予它的任何权限。这一公告发布的前一天，Amazon 要求 Muse 停止使用其电商网站，称 Meta 的 Agent 在该平台购物时未表明自身身份，这引发了关于 Amazon 如何处理客户凭证和账户数据的担忧。

不过，这些银行并未全盘否定智能体商业这一理念，甚至称它可能成为“消费者和商户进行交易的一种主流方式”。

论文写道：“我们同样对智能体商业的前景感到兴奋，并渴望与客户、行业及利益相关方合作，推动其未来发展。”

但银行表示，智能体商业首先需要通过行业标准、政策和消费者保护机制来建立信任，而这些措施目前都落后于技术发展。该联盟目前正在撰写后续论文，详细说明如何实际落实本报告提出的五项关键原则——透明度、安全、隐私与数据、选择权以及互操作性——以确保任何由 Agentic AI 主导的电商部署都尽可能安全。

In a paper published Tuesday, six major global banks, including Bank of America and Capital One, warned that agentic shopping has “risks spanning transparency, safety, privacy & data, choice, and interoperability” that increase “as greater autonomy is given to AI agents.”

AI developers, as well as various retailers and major payment processors, have been working to unveil agentic commerce chatbots, aka AI agents that promise to completely automate the process of online shopping. But adoption has been trickier to quantify. Some consumers use AI chatbots to find stuff or get information on what to buy when shopping online, but many still feel reluctant to let the agents roam free on the internet with access to their entire fortune.

“Consumers are unclear if AI agents will act in their interests,” the report says. “They are concerned that AI agents may buy the wrong thing or spend too much – or even worse, lose their money to scams and fraud.”

Many tech experts also don’t seem to hold total faith in the future of agentic commerce.

Ron Johnson, the former Apple executive behind the original brick-and-mortar Apple Stores, told TechCrunch in a recent interview that he believes AI “will improve the online shopping experience,” but it won’t actually change the “way we shop.”

“AI will never be able to have you physically experience a product,” Johnson said. “They’ll just become more informed shoppers when they come to the store.”

The six banks that co-wrote the report, titled “Building Trust in Agentic Commerce,” seem to agree that many problems and questions still plague this use case of AI agents.

AI agents could prioritize certain products or payment methods simply because of incentives such as higher commissions and lower token costs, rather than choosing what’s best for the customer.

Agentic commerce also has “potential for higher rates of scams, fraud and disputes,” the report says, and any data breach would jeopardize particularly sensitive data belonging to consumers and merchants who rely on these agents.

Case in point: Meta just announced that it found and fixed a zero-day vulnerability in its new AI assistant, Muse, which counts agentic shopping among its promoted capabilities, which could allow an attacker to pretty much hijack the AI assistant and take advantage of any permissions the user had given it. The announcement came only a day after Amazon asked Muse to stop using its e-commerce site, saying that Meta’s agent failed to identify itself while shopping on the platform and that this raised concerns about how Amazon handles customer credentials and account data.

But the banks are not wholesale dismissing the idea of agentic commerce, and even claim that it could potentially become “a mainstream way in which consumers and merchants transact.”

“We, too, are excited by the promise of agentic commerce and are eager to work with customers, industry and stakeholders to enable its future,” the paper reads.

But first, agentic commerce needs to build trust through industry standards, policies and consumer protections, all of which are currently lagging technological developments, the banks say. The consortium is now working on a subsequent paper to detail how the five key principles outlined in this report—transparency, safety, privacy and data, choice, and interoperability—can actually be implemented to ensure that any deployment of agentic AI-led e-commerce is as safe as possible.