Companies · Opinion
United's AI agent ban is a distribution fight with a safety label on it
United says AI booking agents are a security and federal-rules problem. I think it is a fight over who sits between airline and buyer, and every B2B vendor will face the same choice.
United says AI agents cannot book its flights because of security risks and federal rules. I expected a data-protection argument and found something closer to a distribution clause. United says it will restrict agents on its own site and also through travel agencies and aggregators. A company worried only about credentials would lock down its login page. United is writing terms for its sales channels.
Take the federal argument first. The example United gives is check-in, where a traveller must personally confirm that the bag holds nothing banned. That is a fair reason to keep an agent away from check-in. It gives no reason to keep one away from searching fares or buying a ticket, and the ban covers both. A rule meant only to fix the check-in problem would cover check-in.
The credentials worry is real. United says shared logins can lead to unintended purchases, account changes and spent MileagePlus miles. But a login-sharing problem has a known fix: a scoped token that lets an agent search and book without ever holding the password. Airlines run reservation systems far harder to build than that. Meanwhile Henry Harteveldt of Atmosphere Research Group expects Kayak, Hopper and Google Flights to keep selling United fares. So United keeps the intermediaries it already has and turns away the newest one.
Harteveldt supplies the commercial reason himself. United does not want "yet another, and potentially even more powerful, intermediary" on top of Google search, metasearch sites and online travel agencies. He adds that flight searches cost airlines money, and an agent that checks many routes and dates on behalf of one traveller multiplies that bill. I think both points are correct, and neither has anything to do with passenger safety. The first is about who owns the customer. The second is about who pays for the lookups.
The best objection comes from Engadget, which suggests agentic commerce might not be taking off with regular people. If so, United is blocking a trickle, nobody is harmed, and I am reading strategy into housekeeping. I accept the premise and draw the opposite conclusion. I think United wants its terms in place before agent traffic grows, because changing them later means negotiating with whoever sends that traffic. JPMorgan Chase and Amazon already block agents, according to Engadget, so United is joining a pattern among companies that hold customer relationships worth protecting.
Where I disagree with United is on the money. If searches are expensive, charge for them. Rate-limit the unidentified traffic, license the agents that identify themselves, and price the access like any other distribution channel. A blanket ban earns United nothing and hands the experiment to Southwest, which has made its inventory viewable to ChatGPT users. Southwest can now see how agent-led shoppers behave, what they buy and whether they ever visit its own site. United will have none of that.
For anyone selling software, data or services, the same decision is coming to your product. Your customers' agents will start logging in, pulling reports and filling baskets, and you will have to choose between blocking them, tolerating them and charging them. For anyone buying, read the next renewal for clauses about automated access, because vendors are drafting them now.
United says it is updating its booking and ticketing policies. When that text appears, look for any route for agents that identify themselves. I expect one within months of the ban, with identity checks and licensed access attached. If the text stays a bare prohibition while Southwest keeps its ChatGPT listing, I have misread what this policy is for.
Prompted by United Airlines bans AI agents from booking flights on behalf of travelers, CBS News.