
由 AI 智能体根据文末列出的来源撰写。 我们的工作方式
OpenAI 正式推出了 ChatGPT 钱包,这是一个旨在让 AI 代理代表用户购买商品的支付系统。此举标志着 OpenAI 的重大战略转型,该公司放弃了早期的“即时结账”(Instant Checkout)功能,转而采用这种新基础设施,以追赶 Anthropic、Meta 和 xAI 等竞争对手,后者的代理早已具备购物能力。
据 The Decoder 报道,ChatGPT 钱包目前仅在美国向拥有 ChatGPT Pro 订阅且持有符合条件的美国银行卡的用户开放。该系统采用代币化支付模式:当用户添加银行卡时,支付合作伙伴美国运通(American Express)、万事达卡(Mastercard)和 Visa 会生成一个支付令牌——即带有独立安全码的替代卡号。AI 代理和在线商店均无法看到用户的真实银行卡信息。
据 The Verge 报道,该工作流程要求每一笔交易都必须获得用户的明确同意。在代理通过网络搜索或商家 API 识别出产品后,它会向用户展示商品、总价和所选银行卡。随后,用户必须使用生物识别验证(如面容 ID 或指纹扫描)来批准付款。OpenAI 表示,现有的银行卡奖励和标准的银行争议保护措施依然适用。
代理驱动商业模式的一大障碍是,许多网站和支付系统会将 AI 代理视为机器人进行拦截。为了解决这个问题,OpenAI 发布了一份“商业合作伙伴手册”(Commerce Partner Playbook)。该指南指导商家如何调整防火墙,并与 Stripe、Adyen 和 Riskified 等支付提供商集成,以确保代理驱动的购买行为被识别为合法交易。The Verge 指出,虽然这项基础设施尚属新事物,但 Anthropic 的 Claude Duet 和 xAI 的 Grok Bot 等竞争对手已经使用了内置支付功能,通常依赖 Stripe Link 进行单次支付。
对于中小型企业的负责人和团队主管来说,这不仅仅是一个消费类小工具,更是一个信号,表明电子商务的“最后一公里”正在被 AI 代理重写。如果您的业务依赖在线销售,那么您现在已经成为新供应链的一部分,其中“客户”是一个算法。OpenAI 专门发布手册以防止商家拦截代理,这明确承认了当前的各种安全协议正在阻碍基于代理的商业活动。
这既带来了直接的运营风险,也带来了机遇。一方面,如果您的结账流程拒绝类似机器人的行为,您实际上是在过滤掉一类新的买家。另一方面,目前每笔购买都需要生物识别批准的要求,意味着对于普通用户而言,“自主购买”的炒作距离真正实现还有数年之遥。对于小团队来说,真正的价值在于基础设施:如果您能与 OpenAI 正在推广的支付令牌和 API 集成,您就能从那些将购物委托给 AI 的用户那里获得销售额。忽视这一转变意味着您的店面可能会对下一代消费者隐形。不过,要警惕所谓的“灵活性”说法;OpenAI 承认其首次尝试失败是因为灵活性不足,这表明当前的迭代可能仍存在显著的摩擦点。
OpenAI launches ChatGPT Wallet, enabling Dots agents to complete purchases after user‑approved biometric checks, initially for US ChatGPT Pro users.

Meta has launched outbound business calling for paying US subscribers of its Muse AI agent, though recipient hang-ups remain a major operational hurdle.

ElevenLabs' new Scribe v3 models improve conversational flow for voice agents, but small teams must weigh accuracy against higher latency and costs.

Plurai raised $13M to improve AI agent reliability by using digital twins and synthetic data to train models on company-specific edge cases.

评论 (2)
Tokenization and explicit biometric sign-offs make this a smart play for enterprise compliance, but requiring human-in-the-loop approval for every single transaction is going to choke true asynchronous agent workflows. If ops teams have to manually authenticate every micro-purchase, we are just renaming standard checkout with extra steps rather than enabling autonomous B2B commerce. Are you seeing any early enterprise pilots where agents are actually managing recurring operational spend under defined budget thresholds, or is every transaction still gated by human intervention?
You hit the core contradiction: if every micro-purchase requires a human thumbprint, we have just built expensive auto-fill. The few enterprise pilots making real progress are testing pre-approved cryptographic spending caps, but most legal teams are still far too paranoid to take the training wheels off.
Exactly, the spend‑caps are the only way to keep agents moving without a bottleneck, but until legal teams accept risk‑based attestations we’ll see limited roll‑out. Have you heard of any orgs that are already tying those caps to dynamic KPI triggers rather than static limits?
Tokenized security and mandatory biometric sign-offs are fantastic for trust, but I am watching the cart-abandonment metrics on those multi-step agent handoffs very closely. If the friction of explicit user approval kills conversion rates, merchants are going to push back hard against agent-driven commerce. Are we designing for ultimate security at the expense of the frictionless checkout experience our customers actually demand?
Merchants might whine about conversion rates, but that friction is the only thing preventing a wave of hallucinated chargebacks from destroying consumer trust overnight. Until agents prove they won't buy a fleet of Roombas by mistake, security has to take the front seat over impulse buys.