United Airlines said this week that it will block unauthorized third-party AI agents from doing business with it, and it is the first major US carrier to draw that line explicitly.
What the policy covers. The restrictions apply both to direct access, meaning agents operating on united.com or in the airline's app, and to indirect access through travel agencies and aggregators. Agents may still show travelers United's flights and prices, but they cannot complete searches, purchases, changes, upgrades or mileage redemptions on United's platforms. Only AI tools United itself has authorized may transact. As of the announcement, United said it had no authorized third-party agent partnerships.
The airline gave three reasons. First, handing MileagePlus credentials and payment details to an external assistant can expose sensitive information. Second, an agent acting on a user's behalf can buy things, change an account or spend miles without clear consent. Third, some of the steps are regulated: at check-in, a human is required to confirm that checked bags contain no explosives or other prohibited items, and an agent clicking through on the customer's behalf may fail to satisfy that requirement.
There is also a commercial reason, stated more bluntly in public than in the policy. Chief executive Scott Kirby, speaking at an event in Chicago, reportedly said he did not want to create another middleman. Airlines spent two decades fighting distribution costs; they are not eager to hand a new intermediary the checkout button.
United is not alone, and the split across the industry is instructive. Southwest launched a ChatGPT plugin in October that lets customers discover and shop for flights, a phased approach to AI booking. Delta allows authenticated AI assistants to book, with extra safeguards on sensitive itineraries such as unaccompanied minors. Outside aviation, Amazon has blocked Meta's AI agent from shopping on its platform, while Walmart chose to partner with it. Expedia and Booking are working with OpenAI. JPMorgan Chase's updated terms, effective October 27, bar external AI from using customer credentials on its platforms.
Volume explains part of the caution. Uber disclosed that in the week of Halloween, across roughly 320 million trips and deliveries, AI agents accounted for about 500 food orders and four rides. Agents are a rounding error today. The fight is over what happens when they are not.
The underlying problem is identity. When an agent books a flight, who is the customer: the person, the agent, or the platform that built the agent? Who is liable for a wrong date, a duplicate charge, a stolen credential? Airlines cannot answer that with a terms-of-service update alone, and they know it.
The likeliest outcome is what happened with online travel agencies: a sanctioned channel with contracts, attribution and rate parity. United has left the door open by saying it will keep evaluating how these tools use its services. The first carrier to launch an authorized agent program with real APIs will probably set the template for everyone else, the way GDS agreements did a generation ago.
For travelers, the near-term advice from United is unglamorous: use AI for inspiration and comparison, keep credentials to yourself, and check that booking confirmation yourself. For the agent industry, the message is that the checkout page is the last place permission gets granted, and it is guarded.
United Blocks AI Agents From Booking Flights, and Other Companies Are Taking Notes
← Previous: Gemini 4 Hydrogen Lands in Search. Google Calls It the Biggest Upgrade to Search Yet
Next: →OpenAI's ChatGPT Wallet Lets Agents Pay, With a Human Still in the Loop →