United Won’t Let Personal AI Agents Book Flights

Introduction

The moment an AI assistant moves from finding a flight to buying one, the problem changes from search quality to authorization, accountability, and compliance. United Airlines says it will not currently allow personal AI agents to complete ticket purchases for travelers. An assistant may still be able to surface United flights, but the final transaction must be completed through an approved customer flow rather than by an unauthorized third-party agent.

The policy highlights a central tension in the next phase of consumer AI: agents may be technically capable of navigating a booking process, while companies are not yet comfortable giving them the authority to act for customers.

Key points

• Flight information can remain accessible, while booking is restricted. United is not necessarily blocking agents from showing its schedules or fares. It is blocking them from completing purchases or directly interacting with a United account on a customer’s behalf.
• Privacy and unintended actions are central concerns. An agent may handle personal details, payment information, and itinerary data. United is concerned that this information could be exposed, or that an agent could finalize a purchase the customer did not actually intend to make.
• Air travel includes mandatory confirmations. Airlines must provide and collect certain passenger information, including an attestation that a traveler’s bags do not contain explosives. A generic automation layer may not adequately establish that the passenger personally made the required confirmation.
• Other industries are setting similar limits. The material notes that eBay appears to have restricted third-party agents from browsing, while several major banks have also placed limits on agent access. Businesses are increasingly wary of software that can bypass familiar website controls.

United chief executive Scott Kirby said he does not want to create another middleman. The company also says it has not seen many people attempt to use AI agents to book flights, a pattern reportedly similar to what other businesses, including DoorDash, have observed. Yet limited demand does not eliminate the underlying risk. Once an agent can place an order, questions arise over mistaken payments, account exposure, customer consent, and who is responsible when the agent gets something wrong.

Why it matters

For travelers, personal AI assistants may remain useful as search and planning tools, but they are not yet universal digital travel agents. Customers may still need to complete sensitive steps themselves, especially those involving payment, identity, and safety-related declarations.

For companies, the choice is not simply to open or close access. A workable model could require verified agent identities, narrowly defined permissions, explicit approval at critical steps, and records showing what the user authorized. Before an agent can become a transaction layer, a platform must be able to answer four basic questions: who is acting, for whom, whether the user truly consented, and who bears responsibility for an error.

United has not ruled out revisiting its position. It says third-party agents could create opportunities as they improve, but stronger protections for travelers must come first. That suggests the next challenge for agent developers is not just better task completion. It is building authorization, auditability, and accountability into the transaction itself.

Source: The Verge AI