United Airlines said it will block unauthorized third-party AI agents from booking, modifying, or servicing tickets, a move that makes it the first major U.S. airline to draw a hard line against automated shopping agents. The policy, announced Wednesday and reported by CBS News and other outlets, applies to United's website and app as well as to bookings made through travel agencies and aggregators. It covers browser-based agents and extensions, and it applies even when a human supervises or approves each step.
United changed its contract of carriage to enforce the policy. One rule prohibits AI agents from making or modifying tickets or buying services, and another lets the airline cancel reservations made by AI agents without notice, according to PhocusWire. AI agents can still display United flights and fares, but they cannot complete payment or the transaction. Travel agencies may use their own AI tools only with United's explicit permission. Bookings made by AI agents before Nov. 11 will be honored.
The timing is notable. Agentic AI is moving from demonstration to commerce, with assistants that can search, compare, and buy on a customer's behalf. Travel is one of the first large markets where the technology can show whether it works, and United's decision is a test of who controls the interface between airlines and passengers. The airline says it is protecting customers. Its critics say it is protecting its distribution channel.
What United changed
The contract of carriage is the legal document that governs the relationship between an airline and its passengers. United added language that bars AI agents from acting on a customer's behalf for ticketing and related services. The rule is broad enough to cover an agent that logs into a customer's account, an agent that uses a browser extension to fill in a booking, and an agent operated by a travel agency.
The second rule is the enforcement mechanism. It allows United to cancel a reservation made by an AI agent without notice. That is a strong remedy, and it could create problems for passengers who did not know their booking violated the policy. United says it will honor bookings made before the policy took effect, which limits the immediate disruption.
The airline is not blocking AI entirely. Agents can still search for flights and display fares, which keeps United visible in AI shopping tools. What United is blocking is the transaction, the moment when the agent completes a purchase or changes a ticket. That is the point where the airline loses control of the customer relationship and where fraud, privacy, and regulatory risk concentrate.
Why United says it acted
United's stated reasons are security and compliance. Sharing account credentials with an AI assistant can expose sensitive data, enable unintended purchases, or allow unauthorized use of MileagePlus miles. An agent that can log into an account can also change a reservation, add a passenger, or spend stored credit, and the customer may not notice until later.
The regulatory argument is narrower but concrete. Federal rules require a human to acknowledge that a bag does not contain hazardous materials before check-in. An automated agent that checks in a passenger may not be able to provide that acknowledgment. The Department of Transportation and the Federal Aviation Administration have rules about disclosure, refunds, and accessibility that assume a person is on the other end of the transaction. United says its policy helps it comply with those rules.
There is also a simpler operational concern. An AI agent that makes a mistake creates a customer service problem that the airline has to fix. A human who books a flight can be walked through the error. An agent that books the wrong date or the wrong fare may leave the passenger with a nonrefundable ticket and a dispute. United would rather prevent that than resolve it.
The business angle
United's chief executive, Scott Kirby, was blunt about the economics at a Chicago event. "To be honest, I don't want to create another middleman," he said, according to reports. The comment captures the airline's long-running fight with online travel agencies and distribution intermediaries. Airlines want customers to book directly, where they control the experience, the upsell, and the data. Agents that sit between the airline and the passenger threaten that model.
The numbers explain the concern. United reported about $2.1 billion in distribution expenses in 2025, according to reports. Flight searches cost airlines money, and AI agents can generate enormous volumes of searches without booking. An agent that scans hundreds of itineraries to find the best fare can impose costs on the airline without producing revenue. United's policy is partly a way to stop that traffic at the transaction layer.
The market reaction showed that the decision is about more than United. Meta shares fell about 2% after the announcement, according to the New York Post, because the ban undercuts Muse, Meta's AI agent, and raises questions about whether other large consumer companies will follow. Amazon and JPMorgan Chase have taken similar steps to restrict third-party agents. The question is whether United is an outlier or the start of a trend.
The industry split
Other airlines have taken a different approach. American and Delta reportedly will allow some AI tools to complete bookings, and Southwest launched a ChatGPT plug-in for bookings. Those carriers may see agents as a new distribution channel, or they may be experimenting before the market settles. The split means travelers could face a patchwork of rules, with some flights bookable by agents and others not.
The travel industry has seen this movie before. Online travel agencies changed how people booked flights, and airlines spent years fighting over fees, display, and customer data. AI agents are the next version of that fight. The difference is that agents can act autonomously, at scale, and without a clear legal identity. That makes the distribution question harder to resolve with the tools airlines used against OTAs.
The usage numbers are still small. Uber said it handled only four rides and 500 UberEats orders through AI agents during Halloween week, according to reports. United says AI-agent bookings are very few. The policy is therefore anticipatory. United is setting the rules before the volume arrives, which is easier than changing them after agents become a significant channel.
For consumers, the practical effect depends on which agents they use. A general-purpose assistant that books flights directly will fail at United. A travel agency using an approved AI tool may still complete the booking. A traveler who books on United's own site or app is unaffected. The ban is not a ban on using AI to plan a trip. It is a ban on letting an unauthorized agent complete the transaction.
The legal questions
The policy raises questions that courts and regulators have not settled. Can an airline refuse to honor a ticket purchased by an agent acting for a customer? The contract of carriage says yes, but contract terms are not always enforceable, and consumer protection law may limit them. Can an airline cancel a reservation without notice if it believes an agent made it? That could violate DOT rules if the passenger paid and had no idea the booking was prohibited.
There is also the question of authentication. An AI agent that logs into a customer's account looks like the customer to the airline's systems. United's policy asks customers not to share credentials, but many agent products are designed to use them. The same problem has played out at banks, which have been blocking third-party AI agents from logging into customer accounts for similar reasons. A technical solution, such as delegated authorization or a verified agent identity, would be better than a blanket ban, but the standards are still being written.
The broader issue is who bears the risk when an agent makes a mistake. If the agent books a nonrefundable ticket and the customer did not intend to buy, the airline, the agent developer, and the customer may all point at each other. United's policy pushes the risk to the agent and the customer. That may be good for the airline, but it may not be good for the market if it makes agents unreliable or legally risky to use.
What to watch
The first thing to watch is enforcement. A policy in a contract of carriage is only as strong as the airline's ability to detect agents and its willingness to cancel bookings. Watch whether United cancels reservations and how it handles complaints from passengers who say they did not know an agent was prohibited.
The second is the competition. If American and Delta allow agents and United does not, travelers may migrate to the carriers that are easier to book. If agents create fraud or service problems at those carriers, they may follow United. The third is regulation. The DOT could clarify how automated bookings should be treated, and Congress could take up agent liability as part of AI legislation. The fourth is the technology. Verified agent identity and delegated payment could solve the problem United is trying to manage, and the company that builds that standard could unlock the market.
For now, United has made a choice that is defensible on security grounds and revealing on business grounds. It wants to keep the transaction inside its own systems. Whether that protects customers or protects the airline's distribution model will depend on what happens when the agents get better and the volume grows.
Primary sources
- CBS News for United's policy, the contract of carriage changes, and the security and regulatory rationale.
- The New York Post for the market reaction, the Meta share move, and Scott Kirby's comment.
- PhocusWire for the scope of the ban, the travel agency provisions, and the industry context.
- Mavengity's earlier coverage of banks blocking third-party AI agents for the broader agentic-AI context.