United Airlines is restricting unauthorised third-party AI agents from interacting with its digital platforms. Its November 11 announcement reveals a difficult question for the emerging agent economy: how can a consumer safely delegate an action to software when the service receiving that action cannot reliably identify what the customer authorised?

United's own statement, dated November 11, 2026, says the airline is taking steps to restrict access both directly and through agencies and aggregators. This is a move against unauthorised agents, not a claim that every form of AI-assisted travel research is prohibited.

The airline highlights three categories of risk: disclosure of personal account credentials, purchases or account changes that customers did not intend, and use of loyalty-programme miles without appropriate safeguards. It also raises aviation-specific compliance questions, citing the requirement that travellers acknowledge baggage restrictions during check-in.

What United has actually announced

United's stated concern is that a personal assistant acting through a customer's identity may gain broad access to bookings and loyalty accounts without the same control mechanisms. An AI tool using a stored password may be technically capable of logging in, but the airline cannot infer from a successful login that the traveller approved every subsequent transaction.

The Verge's reporting distinguishes the display of United flight options from letting an outside agent complete a booking. That distinction should not be flattened into a claim that United has banned AI from flight search. The publicly described restriction primarily concerns transactions performed by unauthorised intermediaries.

Anadolu's coverage likewise reports United's stated reasons and plans to revise its policies. Neither report establishes the exact technical enforcement mechanism or a universal implementation date. Those details are not documented in the airline's statement.

Why a password is not a delegation contract

The basic problem is the difference between authentication and authorisation. Authentication tells a website which account is being accessed. Authorisation defines what the account holder permits a specific assistant to do, under what conditions, and for how long.

If an assistant signs in using a traveller's credentials, the service may see actions that appear to come from the traveller. But a request to compare fares is not the same as permission to buy a non-refundable ticket, change a reservation, disclose passenger details or use accumulated loyalty points.

This is not evidence that a particular assistant has misused United accounts. It is a structural risk inherent in credential-sharing arrangements. United's statement identifies unintended purchases and account changes as possibilities, not as proven incidents caused by a named product.

The missing agreement between three parties

A workable delegation system would need agreement among the customer, the agent operator and the merchant. The customer could grant narrowly scoped rights: research flights, prepare an itinerary, hold an option or complete a purchase only below a specified price. The merchant would be able to verify those rights without receiving a reusable password. The agent would have a defined identity that allows its actions to be traced and, when appropriate, revoked.

This is our proposed architecture, not an announced United product. Additional safeguards could include expiry times, maximum spend, explicit confirmation for irreversible transactions and transaction receipts that distinguish an agent's action from a direct user action.

Those controls would not eliminate all fraud or errors, but they would give a merchant and customer clearer grounds for investigating a disputed action. They could also reduce the temptation for agent developers to simulate ordinary browser sessions when no supported delegation interface exists.

The counterargument: restriction can block useful innovation

United has legitimate reasons to protect passenger information and ensure compliance with travel rules. But broad restrictions can also deny customers the benefits of comparing many providers, managing complicated itineraries and automating routine travel administration.

Travellers may ultimately prefer independent assistants over a different assistant inside every airline app. That preference would not, by itself, require airlines to give unrestricted access to sensitive accounts. It would, however, strengthen the case for a standardised way to convey user permission.

Some transactions are especially difficult to delegate. A passenger's declaration about prohibited baggage is not merely another checkbox in a purchasing flow. The Verge's reporting describes that example as a reason automated check-in can create regulatory concerns.

What would count as progress?

A meaningful solution would publish a clear distinction between browsing data, creating tentative reservations and finalising transactions. It would let customers revoke an agent's rights and see which steps required their own confirmation. Operators would also need limits on sensitive data collection and clear handling of disputed or duplicated transactions.

The airline's immediate policy is a restriction on unauthorised access. Whether United or the industry will eventually offer interoperable agent permissions remains unknown. The more important signal is that account security and delegated commercial authority are becoming separate product requirements as AI tools move from giving answers to taking actions.

Our assessment: United has identified a real design gap, but a restriction is not the same as a complete answer. The durable opportunity is verifiable, limited and revocable permission—not unlimited credential-sharing, and not a permanent assumption that consumers will only act through one company's own app.

Illustration: original AI-generated conceptual artwork showing an aircraft and an access gate; it is not a photograph of United's systems.

Sources