Travelport and Mastercard: AI agent bought an airline ticket
Travelport, Booked AI, Basis Theory and Mastercard have successfully carried out a real flight ticket purchase via an AI agent, from flight searches through to payment and ticket issuance. The passenger approved the flight and the price, and the agent did not have access to his actual card number.
On 12 November 2026, four companies announced that the entire process had been carried out within a conversation with a digital assistant. A ticket for a major international airline, whose name was not disclosed, was purchased.
For the tourism industry, this example raises the question of how to connect AI assistants with existing distribution and payment systems, and what role travel agencies and airlines would play in such sales.
How did the AI agent buy the plane ticket?
Booked AI used Travelport's TripServices API platform to find the actual flight offer available via the NDC (New Distribution Capability) standard. After the passenger confirmed the itinerary and price, Mastercard Agent Pay enabled the agent-initiated payment. The booking was then processed and the ticket issued through Travelport's and the airline's systems.
In the transaction, the partners had different roles: Travelport provided access to NDC offerings and reservation processing, Booked AI provided the conversational interface and agent, Mastercard handled the payment linked to passenger authorisation, and Basis Theory provided secure storage and tokenisation of card data.
Throughout the transaction, the airline remained the merchant of record, and it processed the payment itself. The demonstration shows one way of using existing NDC and payment systems to sell through an AI interface.
Payment without access to the card number
The card details were stored in Basis Theory's secure system, while the payment agent used a tokenised value instead of the actual card number.
Mastercard's Verifiable Intent technology has linked the passenger's authorisation to a specific booking and amount. According to the partner's announcement, if the transaction details did not match the authorised intent, the payment would not have been processed, and a new authorisation from the cardholder would have been required for any further purchase.
This is an example of the application of the so-called agentic AI: a system that, in addition to providing information, can take actions on behalf of the user within the set permissions.
What could technology mean for agencies?
Agencies today use GDS systems, online booking tools and their own solutions for flight searches, fare comparisons, reservations and ticketing. An AI agent connected to this infrastructure could take over some of the repetitive steps.
For example, a business traveller could instruct the assistant on the destination, budget, preferred departure time, baggage requirements and the employer's travel policy. The agent could then prepare a suitable quote and, once the traveller has approved it, complete the booking. Such an application is potentially of interest to TMCs and online agencies, but this scenario is an illustration of the possibilities, not confirmation of the functionality of any specific test.
For day-to-day operations, post-booking support is also essential: reservation changes, cancellations, refunds, additional services, complex itineraries and flight disruptions. The announcement of the demonstration does not confirm that all of these features are available.
AI agent as a new distribution channel
If travellers make purchases through a conversation with an assistant, the AI interface could become an additional layer between the customer and existing sales and distribution systems. For agencies, it will be important how the agent selects offers, whether they have access to relevant content, and who provides post-purchase support.
Access to the content also depends on the airlines' rules. According to a PhocusWire report, United Airlines has announced restrictions on unauthorised third-party AI agents, including access via agencies and aggregators, citing data protection concerns, unwanted purchases and regulatory compliance.
The technical capability to make a booking therefore does not in itself mean that every agent can access all offers. The relevant agreements, distribution rules and specific integrations will be important. On a related topic, Tragento wrote in its article about AI agents and the distribution of airline tickets.
From a successful transaction to everyday application
The partners have confirmed the actual purchase and issuance of the ticket, but have not indicated when the proposed feature might become widely commercially available. It is also unclear how agencies could contract for and integrate this process into their systems.
The demonstration shows how AI can utilise existing distribution and payment infrastructure. To assess the business value, it will be necessary to see how the solution handles post-purchase changes, liability towards the passenger and a higher volume of transactions.
