美联航禁止未授权 AI 智能体访问,引发 AI 电商控制权争夺
United Airlines Bans Unauthorized AI Agents Starts Battle For Control in AI Commerce
美联航宣布全面限制未经授权的第三方 AI 智能体在其官网和应用中搜寻、购票或管理行程,目前未与任何智能体服务商建立授权合作。Forrester 分析指出,现有身份与访问管理机制难以防范自动化欺诈与流量过载,Amazon 和 Booking 等平台也在收紧限制。分析师建议 B2C 平台仅接入经核准的智能体服务商,并对高风险交易强制执行短期 Token 授权与人工确认。
分析指出了航司封禁第三方智能体背后的风控与控制权考量,为消费端平台构建智能体访问管理提供了可参考建议。
As announced on the United Airlines site, United now restricts all unauthorized AI agents on its site from making material changes to reservations, purchase tickets. Their press release states: “If a third-party AI agent, bot, browser extension or browser feature searches, books, buys, manages or collects information on united.com or the United app for you, and United hasn’t given permission, it’s unauthorized”. United states in its communication that “This data helps us meet regulatory requirements while also enabling us to deliver personalized experiences and improvements to our products and services. This is a responsibility we take seriously.”
United does not yet work with any AI agent providers for authorized AI agent access. It is worth noting that Amazon, Booking.com, EBay, JP Morgan Chase, and Target also have varying levels of restrictions on AI agents accessing their site or customer facing channels.
Organizations like United are restricting unauthorized AI agents for the following reasons:
- Taking proactive measures to own the customer experience – however blocking all AI agent access long term may be counterproductive.
- An org’s existing identity and access management (IAM) and fraud management solutions and layers may not be sufficient to detect and manage AI agents.
- To minimize an org becoming just an AI ‘middlemen’ to broker transactions. Restricting 3rd party AI agents also preserves organizational control over AI agents and avoids AI agents overloading a site with excessive queries that creates service degradation.
- Minimize existing negative experiences. Unvetted and potentially fraudulent AI agents expose companies to additional liabilities and challenges. AI agents have been hacking eCommerce sites at $25 per site.
- Lacking the appropriate defense tooling. While AI agent runtime controls and Know-Your-Agent (KYA) validation are emerging technologies, they may still struggle at scale in prime time, mission critical use.
- Reducing exposure to fines and sanctions. Any loss of personally identifiable information (PII) caused by an AI agent might also negatively impact an organization’s regulatory compliance and brand reputation.
- Waiting out for mature technology and standards. Until liability schemes, mature AI agent verification, and authentication tools with proper human in the loop decisions emerge, many B2C organizations will likely limit 3rd party agent use on their commerce sites.
We recommend that B2C commerce sites 1) define the set of and allow only well known and verified AI agent providers (Anthropic, Meta, OpenAI) to act on behalf of their customers, 2) mandate short term token based AI agent access rights management and delegation, including explicit human performed step-up authentication, and 3) err on the side of caution when customer AI agents try to make high-risk, high-value transactions (such as ticket purchases and reservation changes in United’s case), and 4) audit all agent actions in detail.
Forrester clients who want to dive deeper into this topic and discuss how they should implement Identity and Access Management for customer AI agents can schedule an inquiry or guidance session with me.
Blog
Tokenomics And The AI-Cost Paradox
One of the top discussions today is the AI cost paradox: Why is it that despite falling model prices and rising inference cost-efficiency, AI bills aren’t reducing? A newer, cheaper, more capable model seems to appear every few months. AI startup Reflection released a new open-weight model, Beam, claiming parity with leading Chinese open models […]
Blog
Define Your Business’s Operational Technology Before Attackers Do It For You
Attackers increasingly know which connected systems can disrupt a business, even when defenders don’t. Security leaders must identify and protect critical operational assets before their exposure becomes the next costly disruption.
来源:Forrester Blogs · forrester.com