United Shuts Out AI Agents From Booking Ahead of the Holidays

One of the most intuitive stories about AI agents — "buy me a cheaper ticket" — has just hit its first major wall. According to Bloomberg, United Airlines is blocking third-party AI agents: they can no longer book flights or access customer accounts. The decision is timed to the start of the holiday season, and the stated reason is privacy risk.

An important caveat: the published lede contains no technical details. It is unknown exactly how United is cutting agents off, which agents specifically are affected, or whether there will be exceptions for partners. So what follows is not a retelling of details, but an analysis of what this move means for those building products on top of agents.

What the real problem is

Picture a typical scenario. A user tells an assistant: "Find a ticket to Denver for Friday and buy the cheapest one." The agent opens the airline's website, logs in with the user's credentials, and places the order. From the outside, it looks convenient.

But a personal account holds far more than a single purchase requires: passport data, a loyalty program number, trip history, saved cards. The agent gets all of it at once. If it makes a mistake, is compromised, or simply sends data to the wrong place, the airline sees only "a login with the correct password." Telling a human from a program is hard for it, and the question of liability remains open.

Why airlines are closing up

Privacy is the official argument, and it holds up well. But there are other possible reasons; these are our assumptions, not United's words:

• Control over the channel. The direct website and app are where ancillary services are sold: baggage, seat selection, upgrades, loyalty. If an agent steps in between the customer and the airline, that revenue and the customer data go to a third party.
• Peak load. The holidays are the time of maximum traffic. Automated requests are the last thing anyone wants at such a moment.
• Liability. If an agent buys the wrong ticket or for the wrong date, who handles the refund? It's easier to keep agents out.

What developers and product managers should do

The main takeaway is simple: the "agent browses the site using the user's login" setup is not an integration but a workaround. It works until the site owner decides otherwise. United has shown that it does decide otherwise.

In practice, this means the following:

• Don't build your critical path on browser emulation. If buying tickets is the key feature of your product, dependence on someone else's frontend becomes your biggest risk.
• Look for official channels. These are APIs, partner programs, and delegated access with limited permissions, where the agent can only search and book but cannot read the entire profile.
• Keep a human in the loop. The agent picks options and prepares the order, while final confirmation and payment happen on the user's or the airline's side. That is both safer and more resilient to blocking.
• Read the terms of use before launch. For agents acting on a user's behalf, the legal side is no longer a formality.

Founders should also consider the business risk: if your value is built on access to someone else's platform, that platform can cut off access at any moment, and at the most profitable season for you, no less.

What remains unclear

We don't know whether the ban applies to all agents, including those built into popular assistants, or only to automation from the "outside." We don't know whether United is ready to offer a legitimate alternative, such as its own API for agents. And it is too early to claim that other airlines will follow the same path: some of them may, on the contrary, open their doors to partners and gain a new sales channel.

But the direction is clear. While agents were an experiment, they were tolerated. As soon as they started actually buying on people's behalf, platform owners began talking about privacy, liability, and access rules. Those building agents need to start designing their products for that conversation.