United bans AI booking agents: set your own agent rules now

What United changed

CBS News reports that United Airlines announced on Wednesday 11 November (US time) that third-party AI agents may no longer make reservations for travellers. The restriction applies on United's own website and app. It also reaches, indirectly, into travel agencies and aggregators. The airline says it's updating its booking and ticketing policies to block "unauthorized, third-party personal AI agents or assistants".

The legal change sits in United's contract of carriage. According to One Mile at a Time, the prohibited practices section now includes using AI agents to make, modify or service a ticket. United defines "AI agent" broadly. The definition covers any software that takes autonomous or semi-autonomous action to search for or manage United products, even when a person approves each step. AI-enabled browsers, assistants built into larger platforms and third-party tools that reach United through a travel agency all fall inside the definition.

United gives three reasons, as reported by TechCrunch:

• Customers who give an assistant their login details may expose sensitive information.
• These tools can trigger purchases, account changes or MileagePlus redemptions the customer didn't intend.
• Some US aviation rules assume a human is present, such as the check-in step where a passenger confirms their bags hold no banned items.

United hasn't published an effective date, a detection method or penalties for breaches.

Outside observers see commercial motives as well. Henry Harteveldt of Atmosphere Research Group told CBS News that United wants to keep control of bookings and avoid adding another intermediary after search engines, metasearch sites and online travel agencies. He also pointed to the cost of answering a flood of bot-driven fare searches. Ben Schlappig of One Mile at a Time suspects that United also wants to stop automated repricing and data scraping.

Why a US airline rule matters to NZ businesses

Agents that act for customers are now reaching websites in New Zealand. At TRENZ in Auckland in June, Air New Zealand chief executive Nikhil Ravishankar said the airline was preparing for AI systems that could book travel for consumers. TTG Asia reported that he wants New Zealand and the airline to be "sighted, trusted and machine-readable" by AI platforms. United has gone the opposite way. Southwest, by contrast, has made its flight inventory viewable to ChatGPT users. Two large carriers in the same market have reached opposite answers, which shows the question is still open and depends on each business's commercial position.

Banks are drawing their own lines. TechCrunch notes that Bank of America warned customers in late October not to share their banking passwords with AI assistants. The concern is the same one United raised. A customer who hands credentials to an agent breaks the assumption behind most account security that the person logging in is the account holder.

Few NZ businesses run anything as large as an airline booking engine. Many run something that works the same way, though: an online store, a booking calendar, a B2B ordering portal, a customer account with stored payment details, or a loyalty scheme with points that convert to value. Each of these will receive agent traffic, whether or not the business has decided how to handle it. Most terms of use we see were written before agents existed and say nothing about software acting for a customer. A business with silent terms has made no decision, and its exposure stays unclear until a dispute tests it.

Agents also create practical problems. An agent can place an order the customer didn't mean to place and then ask for a refund. Your fraud rules might flag an agent's login pattern, or they might miss it. The agent can't answer a compliance question meant for a person. Your staff then spend time on disputes your systems were never designed to settle. United has chosen to block agents, and that's one valid answer. A business that wants AI-driven sales might instead offer an approved route, such as a published interface or an app inside an AI platform, with limits on what an agent may do.

The same question applies to your own staff

United's rule also affects businesses on the buying side. Staff are already trying personal agents and AI browsers on supplier portals, travel bookings and software subscriptions. If an employee gives an agent the company's login for a supplier whose terms ban agents, the business carries the breach. The supplier may also be entitled to cancel the transaction. One Mile at a Time reports that United's wording is broad enough to cover award search tools. Schlappig asks whether United will enforce the rule against all of them or only some. Your staff can't predict that enforcement either, so a written internal rule beats guessing.

Practical next steps

• List your exposure points. Record every customer-facing system where someone can log in, pay, book or redeem value. Note which ones store payment details or credentials, because agents will reach those systems first.
• Pick a position for each system. Choose whether to block agents, allow them or offer a controlled route. A booking calendar for a local service business and a trade portal with credit terms will likely need different answers.
• Update your terms of use. Define what you mean by an AI agent and state what an agent may and mayn't do on your systems. Ask your lawyer who bears the cost of an unintended purchase. United's definition is a useful reference because it includes semi-autonomous tools where a person approves each step.
• Check your logs and fraud rules. Look for automated sessions behind customer logins and find out whether your current controls notice them. Treat a rise in high-volume searches or unusual login locations as a prompt to investigate.
• Write an internal agent policy. Tell staff which agents they may use, which company accounts those agents may touch and whether credentials may ever be shared with them. Make supplier terms part of the approval check. Our AI consultancy work starts with this kind of governance.
• Plan an approved route if you want agent sales. If customers buying through AI tools suits your business, build a deliberate interface with spending limits and confirmation steps. Don't leave agents to scrape pages built for people. Our AI agents page covers how we approach agent design and controls.

United's rule is a single airline's commercial decision, and other carriers have gone the other way. The decision it forces is broader. Every business with a login page will need a stated rule for software acting on a customer's behalf, and writing that rule now costs far less than sorting out the first disputed transaction.