This piece is based on United’s statements, its revised contract of carriage as reported by the travel and technology press, and comparable policies from other carriers. I have not tested any assistant against United’s systems, and claims are attributed to the outlets that reported them.
AI booking agents just hit their first real wall. United Airlines said in November 2026 that it will restrict unauthorised third-party personal AI agents from booking or changing its flights, on its own website and app and through travel agencies and aggregators. Your assistant can still find United fares. It cannot buy them.
My honest take: United is partly right and partly protecting its margins, and both of those things will shape how every AI booking agent works over the next year. If you use an assistant to plan trips, nothing much changes for searching. If you build agents that transact, this is the week the rules of the road got written, and they were not written by you.
Table of Contents
- What United Did to AI Booking Agents
- Why United Says It Is Blocking AI Booking Agents
- Lesson 1: Read the Contract, Not the Headline
- Lesson 2: AI Booking Agents Still Win at Search
- Lesson 3: Never Hand an Agent Your Login
- Lesson 4: This Is a Fight About the Middleman
- Lesson 5: Authenticated AI Booking Agents Are the Likely Compromise
- Lesson 6: Builders Need Identity, Not Workarounds
- What Happens Next for AI Booking Agents
- How Airlines Treat AI Booking Agents Right Now
- FAQ About AI Booking Agents
- Verdict
- Sources and Further Reading

What United Did to AI Booking Agents
United’s statement, as quoted by Skift, says the airline is restricting access directly within its digital platforms and indirectly through travel agencies and aggregators, to prohibit use by unauthorised, third-party personal AI agents or assistants. That covers the obvious route, an agent logging into united.com, and the less obvious one, an agent booking through an online travel agency on your behalf.
The ban sits in the contract of carriage, the legal document that governs every ticket. According to Silicon Report, the revised version defines covered tools as software taking autonomous or semi-autonomous action, including consumer platform assistants and AI-enabled browsers. That second phrase matters. It is not only chatbots. A browser with a built-in agent that fills in the booking form counts too.
The penalty is not trivial. Business Traveller reports that United reserves the right to cancel reservations made through prohibited tools without notice. So an AI booking agent that gets through the front door can still leave you with a cancelled ticket at the gate, which is a far worse outcome than a blocked purchase.
What remains allowed is search. Per The Wall Street Journal, as summarised by Digital Trends, assistants can still surface United’s flight options and fares. They just cannot complete the transaction themselves.
Thrifty Traveler names xAI’s Grok bot and Meta’s Muse as examples of the kind of assistant the rule targets, and calls United the first US airline to publicly block AI agents from booking flights.
Why United Says It Is Blocking AI Booking Agents
United gives three reasons, and they are worth taking seriously before you dismiss this as protectionism.
Account security. Most AI booking agents work by logging in as you. Handing a MileagePlus login to an outside tool exposes the account, the stored payment details and the miles balance to whatever that tool does, or whoever compromises it.
Unintended actions. An agent could buy the wrong flight, change a reservation you meant to keep or spend miles you were saving. Anyone who has watched an agent misread a date picker knows this is not hypothetical.
Aviation rules. This is the argument I had not seen before. Thrifty Traveler’s example is check-in: a passenger has to confirm their bags contain no prohibited items. An AI booking agent cannot truthfully make that declaration for you. United questioned whether automated actions comply with FAA and DOT rules at all.
Those are fair points. They are also convenient ones, which brings us to the lessons.
Lesson 1: Read the Contract, Not the Headline
Most coverage framed this as “United bans AI”. The contract is narrower and, in places, broader than that. It does not stop you using ChatGPT, Claude or Gemini to plan a trip. It does cover AI-enabled browsers, which many people would not think of as agents at all.
If you travel often, the practical rule is simple: research with whatever assistant you like, then complete the booking yourself on United’s site or app. If you use an agentic browser, switch off its autonomous mode before you reach the payment page.
For businesses, the lesson is broader. Contracts of carriage, terms of service and acceptable use policies are where companies are now writing their rules for AI booking agents. Legal teams that have not read their own terms with agents in mind should do so this quarter.
Lesson 2: AI Booking Agents Still Win at Search
This is the part that gets lost. United did not block discovery, and discovery is where assistants are genuinely good. Comparing departure times across three airports, checking which fares include a carry-on, spotting a cheaper routing through a different hub: that is exactly the tedious work AI booking agents do well.
Visa’s new agentic commerce benchmark, published the same month, found that models failed 42.8% of shopping runs, mostly by buying before they had uncovered what the customer needed. Search-then-human-checkout is not a compromise. On current evidence, it is the right design.
So use the agent for the first 90% of the job. Ask it for three options with the reasons each one fits, then book the winner yourself. You lose a few minutes and avoid a cancelled ticket.
Lesson 3: Never Hand an Agent Your Login
United is not alone in worrying about this. Bank of America urged customers the week before not to share their passwords with AI assistants, according to reporting on the United move. When an airline and a major bank deliver the same warning within days, it is worth listening.
The underlying problem is that most AI booking agents today act as you, with your credentials, rather than as a clearly identified agent with limited permissions. That means a site cannot tell the difference between you and the bot, and neither can its fraud systems. If something goes wrong, the paper trail says you did it.
Until airlines offer proper delegated access, the safe position is that no third-party agent gets your primary login. If an assistant asks for it, decline. The same goes for agents connected to your email, where a single injected instruction can do real damage, as we explain in our guide to prompt injection and agent security.
Lesson 4: This Is a Fight About the Middleman
Security is the stated reason. Distribution is the unstated one. The day before the announcement, United CEO Scott Kirby said he did not want to create another middleman, according to Thrifty Traveler. Airlines spent two decades pulling customers away from travel agencies and onto their own sites, where they control pricing, upsells and loyalty.
AI booking agents threaten all of that. An assistant that picks the itinerary and buys the ticket gains enormous influence over the sale, as Business Traveller points out, much like the comparison sites and agencies airlines have fought for years. Thrifty Traveler’s author goes further and argues the bigger threat to United’s profits is agents rebooking flights automatically when cheaper fares appear.
That is opinion, not fact, and United says it is not currently seeing many agent bookings. But you do not need much imagination to see why an airline would want to set the rules before agent traffic grows, rather than after.
Lesson 5: Authenticated AI Booking Agents Are the Likely Compromise
Not every airline has chosen United’s path. Delta says AI assistants can complete bookings once they are authenticated, with extra safeguards for sensitive transactions such as trips for unaccompanied minors. Southwest is not blocking agents and has launched a ChatGPT integration that lets travellers search its flights inside the chatbot.
Delta’s approach is where I expect the industry to land. An authenticated agent is one the airline knows about, can rate-limit and can hold accountable. It solves the security problem without handing the customer relationship to a third party, because the airline decides who gets authenticated and on what terms.
United’s wording leaves room for this too. It prohibits unauthorised agents. That word implies an authorised route may come later, most likely through partnerships the airline chooses and controls.
Lesson 6: Builders Need Identity, Not Workarounds
If you build AI booking agents, the temptation will be to route around the block: rotate user agents, spoof browser fingerprints, book through an aggregator that has not caught up yet. Do not. Bookings made that way can be cancelled without notice, and your users will blame you, not United.
The better path is identity. The card networks are already building it. Visa’s Intelligent Commerce includes a directory of verified agents that merchants can trust, and Mastercard’s Agent Pay uses agentic tokens to show a transaction came from an approved agent. Airlines will want something similar before they open the door.
Design your agent so that it can identify itself honestly, hand checkout back to the user cleanly and respect a site’s stated policy. Our guide to AI browser agents covers how the main tools handle checkout handoffs today, and building an agent from scratch shows where to put those guardrails in your own code.
What Happens Next for AI Booking Agents
Bloomberg framed the United move as coming ahead of holiday travel, and the timing makes sense. The busiest booking weeks of the year are exactly when an airline least wants a wave of automated purchases, changes and cancellations hitting its systems from tools it cannot see.
Three things are likely over the next six to twelve months. First, other carriers will publish their own positions, and they will split roughly along United and Delta lines: block unknown agents, or authenticate them. Second, the big assistant makers will push for formal partnerships, because an agent that cannot book is only half a product. Third, hotels, rental car firms and ticketing sites will watch the airlines closely and copy whatever survives the holiday season.
The regulatory angle is the wild card. United raised FAA and DOT compliance as a reason for caution, and regulators have not yet said how they view automated declarations at check-in or automated changes to bookings. Clear guidance in either direction would change the economics for every company building AI booking agents.
There is also a consumer protection question nobody has answered. If an agent books the wrong flight, who is responsible: the traveller, the agent maker or the airline that accepted the booking? Card networks are trying to answer that with verified agent identity and token controls. Airlines are answering it, for now, by saying no.
My bet is that within a year, the main assistants will be able to book with at least one major US airline through an authorised route, with spending limits and confirmation steps baked in. The open web, where an agent simply logs in as you, is closing. The approved lane is just starting to open.
How Airlines Treat AI Booking Agents Right Now
The picture is still moving, but here is where the major US carriers stood in mid-November 2026, based on published reporting.
| Airline | Stance on AI booking agents | What it means for you |
|---|---|---|
| United | Blocks unauthorised personal agents from booking or changing flights, including via agencies | Search with an agent, book yourself |
| Delta | Allows bookings by authenticated assistants, with extra safeguards for sensitive trips | Agent checkout may work through approved routes |
| Southwest | Not blocking; offers flight search inside ChatGPT | Agent-friendly discovery, booking still on Southwest |
| American | No public position reported | Check the current terms before relying on an agent |
For travellers, this table reduces to one habit. Check the airline’s terms before letting any agent pay, and book directly whenever you are unsure.
| If you are | Start with | Why | Link |
|---|---|---|---|
| A frequent flyer | Agent search, direct booking | Avoids cancellations and keeps your loyalty account private | United contract of carriage |
| A travel manager | Update your travel policy | Staff using agentic browsers may breach airline terms without knowing | Business Traveller report |
| An agent developer | Human checkout handoff | Bookings by blocked tools can be cancelled without notice | AI browser agents |
Related on this site: for more context, see Siri AI Two Months In and Claude Haiku 5.5.
FAQ About AI Booking Agents
Can I still use ChatGPT or Claude to plan a United trip?
Yes. Assistants can still surface United’s flights and fares. The restriction applies to completing a booking or changing a reservation on your behalf.
What happens if an AI booking agent books a United flight anyway?
United reserves the right to cancel reservations made through prohibited tools without notice. That is the main practical risk, and it is a good reason to book yourself.
Does the ban include AI browsers?
According to reporting on the revised contract of carriage, yes. Covered tools include consumer platform assistants and AI-enabled browsers that take autonomous or semi-autonomous action.
Why would an airline block AI booking agents?
United cites account security, unintended purchases or changes, and doubts about whether automated actions comply with aviation rules such as check-in declarations. Analysts also point to the airline’s interest in keeping customers on its own channels.
Do other airlines allow AI booking agents?
Delta says authenticated assistants can complete bookings with extra safeguards, and Southwest offers flight search inside ChatGPT. Policies vary and are changing quickly.
Is it safe to give an AI agent my airline login?
Not today. Both United and Bank of America have warned against sharing credentials with assistants. Use agents for research and keep logins and payment in your own hands.
Verdict
Keep using AI booking agents for what they do best: searching, comparing and shortlisting. Book United flights yourself, on United’s site or app, and switch off autonomous mode in any agentic browser before you reach checkout. That avoids the one real risk here, a cancelled reservation.
The caveat is that United’s security case is genuine but incomplete. The airline also has every commercial reason to keep agents from becoming a new middleman, and it will likely allow authorised partners on terms it controls. Expect “unauthorised” to do a lot of work in the months ahead.
Your next step: if you manage travel for a team, add one line to your travel policy this week saying staff must complete airline bookings themselves, not through autonomous agents or AI browsers. It costs nothing and prevents an awkward conversation at a departure gate.
Sources and Further Reading
- Skift’s report, for United’s own statement on restricting agents directly and through agencies
- Business Traveller, for the cancellation clause and the middleman argument
- Silicon Report, for how the contract of carriage defines covered tools
- Digital Trends, for the WSJ detail that agents can still surface fares
- Thrifty Traveler, for the FAA and DOT argument, the Delta and Southwest comparison and the rebooking theory
For more on this site, see how agents navigate real websites in AI browser agents, how to keep them from being hijacked in prompt injection and agent security, and how to run one privately on your own machine in local AI agents. For phone-based assistants, read AI voice agents.

