2026年4月17日 / 美国东部时间凌晨5:00 / 哥伦比亚广播公司新闻（CBS News）
作者：梅根·塞鲁洛（Megan Cerullo）
梅根·塞鲁洛是驻纽约的CBS财经观察（CBS MoneyWatch）记者，报道小企业、职场、医疗保健、消费支出和个人理财话题。她经常做客哥伦比亚广播公司新闻24小时频道解读自己的报道。
查看完整简介
人工智能“代理”承诺可以包办一切：从像近藤麻理惠整理衣橱那样帮你清理收件箱，到根据你的预算和风格偏好为你挑选一双高跟鞋。
但科技专家警告称，将关键决策外包给AI会让消费者面临风险，可能导致沟通失误、造成财产损失，还可能让黑客获取用户的数据权限。在所谓的“代理式商务”，也就是依靠AI代理为你购物时，这种风险尤其突出。
“这项技术尚未普及，目前风险相当高，因为系统中还没有足够的保障措施，让消费者放心让代理自主为他们购物，”波士顿咨询集团的AI专家马特·克罗普（Matt Kropp）告诉哥伦比亚广播公司新闻。“它有可能帮你买车，但我绝不会说‘把我的信用卡给它’。”
崭新时代
尽管存在这些担忧，美国一些大型企业仍在推进AI商务业务，它们将其视为吸引客户、通过让AI为消费者跑腿来推动更多销售的新方式。
例如，美国运通（American Express）本周宣布为使用指定AI代理进行消费的持卡人推出新服务和保障措施。这家信用卡发卡机构表示，其中包括验证代理在消费时的身份。美国运通在一份声明中称，这项服务“将保护符合条件的客户免受AI代理失误引发的扣款纠纷”。
亚马逊的AI代理助手名为“鲁弗斯（Rufus）”，可以追踪这家在线零售平台上的商品价格，在价格达到用户设定的阈值时提醒消费者，并完成下单支付。
美国最大零售商沃尔玛已经推出了名为“斯帕基（Sparky）”的“对话式”AI代理，该公司表示，这款代理可以帮助消费者查找商品、提供用户评价并协助下单。
市场研究机构Statista去年11月的数据显示，约四分之一的18至39岁美国人曾尝试使用AI研究产品或购物。
可能出现哪些问题？
AI的加速普及也引发了诸多失误案例。
以旧金山一家科技初创公司的创始人塞巴斯蒂安·海涅曼（Sebastian Heyneman）的经历为例。据《纽约时报》报道，他指示一名AI代理帮他争取在瑞士达沃斯世界经济论坛上的演讲机会。这款聊天机器人成功帮他拿到了这个备受追捧的年度权贵聚会席位——但收费3万美元，这超出了他的承受能力。
海涅曼使用的是Tasklet公司的聊天机器人，该公司允许企业用AI代理自动化日常商务任务。Tasklet创始人安德鲁·李（Andrew Lee）告诉哥伦比亚广播公司新闻，当用户的提示给AI下达了相互矛盾的指令时，就可能出现这类问题。
李还表示，如今的代理式AI完全有能力为消费者购物，完成“消费者日常能做的普通事务”。但他警告称，仅仅因为技术能够实现某项功能，并不意味着就应该以这种方式使用它。
“就目前而言，购物这个特定用途并不适合使用这些系统，”他告诉哥伦比亚广播公司新闻。“这些代理本质上很难让人信任。就我个人而言，我还不太放心这种方式。我更喜欢自己掌控资金的去向，作为企业，我们也不建议客户这么做。”
原因在于，恶意行为者可以诱骗AI代理泄露消费者的个人信息，纽约一家科技初创公司的创始人布雷顿·奥尔巴赫（Bretton Auerbach）告诉哥伦比亚广播公司新闻。
“如果你把信用卡交给代理，说‘去这个网站帮我在线买东西’，就有办法骗过这个代理，”他说。“它可能会把一个合法网站误认为是钓鱼网站，而那个钓鱼网站会用醒目的大字写着‘在此粘贴你的信用卡号’。”
编辑：阿兰·谢尔特（Alain Sherter）
AI “agents” can do your shopping. Should you let them?
April 17, 2026 / 5:00 AM EDT / CBS News
By Megan Cerullo
Megan Cerullo is a New York-based reporter for CBS MoneyWatch covering small business, workplace, health care, consumer spending and personal finance topics. She regularly appears on CBS News 24/7 to discuss her reporting.
Read Full Bio
Artificial intelligence “agents” promise to do everything from tidying up your email like Marie Kondo declutters a closet, to buying you a pair of heels based on your budget and style preferences.
Yet technology experts warn that outsourcing key decisions to AI exposes consumers to risks, potentially leading to communications errors and costing people money, while also potentially handing hackers the keys to their data. This is particularly true when it comes to so-called agentic commerce, or relying on AI agents to make purchases for you.
“It isn’t mainstream yet and it’s pretty risky right now, because there aren’t enough guardrails in the system for people to feel comfortable with agents autonomously buying things for them,” Matt Kropp, an AI expert with Boston Consulting Group, told CBS News. “It could potentially go buy a car, but I wouldn’t say, ‘Here’s my credit card.’”
Brave new world
Such concerns aren’t keeping some of America’s biggiest companies from charging ahead with AI commerce, which they see as a new way to engage customers and to drive more sales by letting AI do the legwork for shoppers.
For example, American Express this week announced new services and protections for cardholders who make purchases using specified AI agents. That includes verifying the identity of an agent when it makes a purchase, according to the credit card issuer. The service “will protect eligible customers from charges related to AI agent error,” Amex said in a statement.
Amazon’s agentic AI assistant, dubbed “Rufus,” can track the price of products on the online retailer’s platform, alert customers when the price hits a prescribed level and complete the purchase.
Walmart, the biggest U.S. retailer, has deployed what it calls a “conversational” AI agent named Sparky that the company says can help consumers find products, provide customer reviews and help with ordering.
Roughly a quarter of Americans between the ages of 18 and 39 say they have tried using AI to research products or to shop, according to November data from market research firm Statista.
What could go wrong?
The accelerating adoption of AI is also leading to mishaps.
Consider what happened to Sebastian Heyneman, the founder of a San Francisco-based tech startup. According to the New York Times, he instructed an AI agent to secure him a speaking opportunity at the World Economic Forum in Davos, Switzerland. The bot succeeded in landing him a coveted slot at the annual gathering of powerbrokers — for $30,000, a fee he couldn’t afford.
Heyneman used a bot by Tasklet, a company that lets businesses automate routine business tasks with AI agents. Andrew Lee, the founder of Tasklet, told CBS News that such problems can arise when a user prompt gives the AI conflicting instructions.
Lee also said agentic AI today is fully capable of shopping for people and doing “normal things consumers can do.” But just because tech can do something doesn’t mean it should be used in that fashion, he warned.
“The specific use case of shopping is not a good thing to use these systems for — yet,” he told CBS News. “The agents are fundamentally hard to trust. Personally, I am not super comfortable with that yet. I like to control where my money goes myself, and as a business, we don’t recommend that.”
The reason: Bad actors can lure AI agents into turning over a consumer’s personal information, Bretton Auerbach, founder of a New York-based tech startup, told CBS News.
“If you give an agent your credit card and say, ‘Go to this website and buy me something online,’ there are ways to trick the agent,” he said. “It might mistake a legitimate website for a phishing website that says in big, bold, text, ‘Paste your credit card number here.’”
Edited by Alain Sherter
发表回复