Tarun Upaday.

Operating Notes

Whose login is your AI agent using?

On November 11, United Airlines said it is restricting access for unauthorized, third-party personal AI agents and assistants. The restriction covers United’s own app and website, and indirectly travel agencies and aggregators, through changes to its booking, ticketing and distribution policies.

The Hacker News discussion ran to almost 700 comments, and most of them argued about motive. Some readers think United wants to protect fare complexity that rewards people who don’t shop around. Others doubted that anyone really wants an agent to book their flights. Several pointed out that an agent driving an ordinary browser is hard to tell apart from a person. Those are opinions, and I can’t verify United’s motive either.

What I can read is the reason United gave. It named two. Travelers who hand their login to an assistant expose their data, and the assistant can make purchases, change the account or spend MileagePlus miles the traveler didn’t intend. And check-in requires a person to confirm that their bags hold no banned items, under FAA and DOT rules, which an agent can’t honestly do for someone else.

Both reasons come down to the same question: who is acting, and on whose authority? That’s the part I’d take seriously, whatever you think of the motive.

Borrowed identity is the weak point

Many consumer agents work by borrowing the person’s identity. You give the agent your password or your logged-in browser, and to the airline it looks like you. That’s convenient until something goes wrong. Then the airline sees a purchase from your account, and nobody can tell which decisions were yours.

My own AI agents can research a booking, but they can’t make one without my approval. That’s a rule I set for myself, not one a supplier imposed. United is now making a similar point from the other side of the counter.

Corporate travel has worked this out before. A travel management company books with its own agency credentials, under the company’s travel policy, using a traveler profile the company controls. The airline can see that an agency is acting, and for whom. An AI agent in that setting should work the same way: its own identity, a defined scope, and a person who approves anything that spends money.

United’s release doesn’t say what counts as an authorized agent, when the changes take effect, or how agency bookings will be treated. It says its Digital Trust Center will explain more. Until it does, I’d treat any agent that books on a traveler’s personal login as exposed.

A worked example

Take a hypothetical. A sales director’s 6 p.m. flight out of Denver is cancelled. What should a corporate travel agent do?

It can notice the cancellation from the reservation feed, before the traveler opens the airline app. It can check the company’s travel policy, find two rebooking options, and hold both if the fare rules allow it. Then it messages the traveler with the options and a deadline to choose.

It should not log in as the traveler. The rebooking goes through the agency’s own channel, under the company’s account, after the traveler picks an option. If the new fare costs more than policy allows, a travel manager approves the difference.

Then it confirms the outcome. The new ticket is issued, the old one is refunded or held as credit, and the hotel and car match the new arrival time. A task like this isn’t done until the world has changed. Check-in stays with the traveler, because the statement about what’s in their bag is theirs to make.

Every step that touches inventory or money happens under an identity the airline can see and a person can audit. That record also matters when two agents end up competing for the same scarce inventory.

What to ask this week

If you run travel for a company, ask your agency and every AI tool vendor one question: when your agent books, whose credentials does it use? If the answer is the traveler’s own login, expect that path to break, starting with United.