United Airlines has told unauthorised third-party AI agents to stay out of its booking systems. In an announcement dated 11 November 2026, the airline said automated personal assistants could expose passenger accounts, change reservations or make purchases that a customer never intended. Its new restrictions cover direct digital channels and access through travel agencies and aggregators, raising new questions about AI flight booking. They do not amount to a ban on every use of AI in travel.
The official United statement makes the airline’s reasoning unusually clear. A customer can ask a chatbot to compare flight options, but allowing that chatbot into an account is a different proposition. An assistant with credentials may be able to spend money, redeem loyalty points and alter an itinerary. The concern is not that an AI can discuss flights. It is that an unauthorised agent might take actions for which a traveller remains responsible.
Why United is restricting AI flight booking
Imagine asking a personal assistant to find the cheapest way to travel and discovering that it has chosen a poor connection, paid for an unnecessary extra or used miles you intended to save. Those outcomes would be frustrating even if a human travel agent had made them. They are harder to untangle when a software agent was given broad permission and its instructions were never fully understood by either side.
United says passengers sharing account credentials with outside AI assistants may expose sensitive personal information. It also highlights the possibility of unintended purchases and account changes. This is a practical security problem, not merely a philosophical argument about who should control future travel. Booking a flight involves names, payment details, travel plans and sometimes loyalty account balances. A mistake can travel well beyond an inconvenient chat response.
The airline has also identified an aviation-specific complication. Some check-in steps require a traveller to acknowledge safety statements personally, such as confirming that baggage does not contain prohibited items. United cites United States aviation rules in explaining why an automated tool should not simply tick such boxes on behalf of a passenger. A program may know which answer is normally selected without being the person who packed the bag.
The restriction is narrower than a complete AI ban
The wording matters. United says it is restricting unauthorised third-party personal AI agents, including through travel intermediaries. That is not the same as announcing that its own customer service technology can never use AI or that ordinary customers are prohibited from consulting a chatbot. Neither does the statement prove that every automated booking service is already blocked on every website around the world. The action is a policy and access restriction whose practical effects will depend on implementation.
Travel-industry publication PhocusWire reported on the airline’s approach, putting the move in the context of airlines and online booking platforms reassessing automated access. The distinction between a recognised commercial distribution partner and an unknown personal agent is central. A travel business using authorised systems operates under arrangements that a consumer chatbot may not have.
United says its policies are being updated across booking, ticketing and distribution. That is a significant point for software developers tempted to treat a public-facing website as an invitation to automate anything a person could click. Access rules, authentication and safety responsibilities still apply when software is operating the browser. A tool being technically capable of an action does not mean the operator has permission to perform it.
Booking a holiday is more than finding a cheap fare
For a traveller, the appeal of agent-led booking is obvious. A single request could in principle compare dates, work out connections, reserve seats and manage a change when plans unravel. LiveAIWire has examined Airbnb’s plans for AI-assisted travel booking, where the attraction similarly lies in moving from suggestions to action. The difficulty is that the moment a system starts spending money or accepting terms, helpful advice becomes delegated authority.
The same issue underlies the trust gap in AI holiday planning. People may welcome a tool that produces a shortlist but still want to review the final itinerary and price before anything is purchased. An airline ticket can carry restrictions on changes, baggage or refunds that are not obvious from a headline fare. An AI agent would have to understand those details and explain them correctly rather than simply optimise for the first visible price.
There is also a question of how a mistake is corrected. If an automated assistant books the wrong date, who has authority to cancel it? Can an airline recognise the customer’s instruction as genuine? Does the customer know exactly which permissions were granted, and can they be revoked? These are not reasons to abandon automation altogether. They are reasons to build clear boundaries before a software agent is given access to money or accounts.
What travellers should do now
United customers need not stop using AI to research destinations, compare routes or prepare questions before booking. The safer distinction is to keep research separate from account access until a tool is explicitly authorised through a recognised channel. That means avoiding the casual sharing of passwords, checking the final itinerary personally, and being especially cautious about a third-party service promising to complete every step without a meaningful review.
A booking confirmation is not the place to discover the difference between advice and agency. A passenger should know who took the payment, which company holds the reservation, and whom to contact if something changes. When an assistant can explain alternatives but cannot safely complete a purchase, it is better to say so than to create a false impression of completion.
United’s decision illustrates a boundary that other industries may increasingly have to draw. Customers want the convenience of an AI assistant that gets things done, while companies have legitimate reasons to protect accounts and require human agreement to sensitive actions. The airline has chosen to put that boundary in writing. Whether its restrictions become a model for others will depend on what safer, clearly authorised agent systems can demonstrate next.
About the Author
Stuart Kerr is Technology Correspondent at LiveAIWire, covering artificial intelligence, cybersecurity and the social impact of emerging technology. LiveAIWire is an independent, human-led technology publication using AI-assisted research, editorial production and original AI-assisted editorial illustrations under his direction.