Skip to main content
Back to News
analysis/AI Infrastructure

United Blocks AI Agents, and the Agent Internet Splits in Two

United Airlines will block third-party AI agents from booking flights and manage reservations, pushing the restriction down to travel agencies and aggregators.

Stefan Trbojevic

Stefan Trbojevic

12 November 20266 min read
LinkedIn
Dark airport terminal concourse with a single lit gateway in the distance

The takeaway

Credential-sharing integrations are being closed off one institution at a time. Agents that can sign their own requests and work through contract channels survive; agents that log in as the customer will keep losing access.

Why it matters for builders

If your agents touch other companies' systems, stop treating the customer's password as an integration. Ask every vendor who signs for your agent, prefer merchant-hosted APIs and MCP servers over browser automation behind a login, publish an RFC 9421 key directory so you can be verified the moment a site enables Web Bot Auth, keep a human in the loop where law requires a human act, and log consent plus every action so liability arguments hold up.

United Blocks AI Agents, and the Agent Internet Splits in Two

United Airlines said this week that it will restrict AI agents from booking flights and managing reservations on customers' behalf, citing cybersecurity risk and the possibility of breaking federal aviation rules. The statement is short and bureaucratic, but it is the clearest sign yet that the agentic web is splitting into two tiers: companies that have a negotiated lane for agents, and companies that will simply close the door.

What United Actually Did

The carrier is not patching a bug. It is changing policy. "United is taking steps to restrict access, directly within our digital platforms and indirectly via travel agencies and aggregators, to prohibit use by unauthorized, third-party personal AI agents or assistants," the airline said, adding that the work "includes updates to our booking, ticketing and distribution policies," according to CNBC.

Three concerns are named. First, credential sharing: handing an assistant your MileagePlus login can expose sensitive account data and produce unintended purchases, account changes, or drained miles. Second, regulatory exposure: flight check-in requires a human to acknowledge that bags contain no prohibited items, an attestation an autonomous agent cannot legally make. Third, control: the airline wants to decide who transacts on its systems.

The phrase "indirectly via travel agencies and aggregators" is the part builders should read twice. United is not only blocking traffic at its own front door; it is pushing the restriction down the distribution chain, so an agent that works today through one booking path may fail tomorrow through another, with no code change on the builder's side.

The Root Cause Is Missing Identity

Airlines and banks are not fighting agents. They are fighting an authentication model that has no slot for them.

Today the only portable credential a consumer agent can use is the user's own password. That makes three actors look identical to the server: the real customer, a helpful assistant acting for them, and an attacker holding a credential dump. When an institution cannot separate those three, blocking all of them is the rational default.

Banks got there first. In late October, Bank of America emailed customers under the subject "Protecting your finances in the AI World," urging them never to share a username and password with another person or an outside service, "including an AI-enabled tool or agent that offers to act on your behalf." The email also set the liability default: "If you do share your credentials, keep in mind that you're responsible for activity resulting from that access," TechCrunch reported.

The agent has no identity of its own, the human carries all the risk, and the institution has no way to audit intent.

Two Camps, One Fault Line

The industry response is already splitting, and the split is not about technical sophistication. It is about whether a company has a contract channel for agents.

Companies with a partner lane are opening up. Expedia and Booking are working with OpenAI on agent-driven travel. Walmart has partnered with Meta's shopping agent Muse. In each case there is a commercial relationship, an API surface, and someone to call when something goes wrong.

Companies without one are closing. Amazon blocked Meta's Muse. eBay quietly shut its listings to third-party AI agents last week, as we covered here. United now joins them, and its policy explicitly reaches into the aggregator layer where no direct relationship exists.

This is the scraping fight of 2023 to 2025 repeating: the sites that could contract bot traffic built interfaces, and the rest built walls. Agents that depend on credential sharing sit on the wrong side of that wall by construction.

A glowing glass cylinder with a cyan core, standing for an agent identity that can be verified

The Fix Already Exists. The Airlines Have Not Adopted It.

The technical answer to "is this really the customer's agent?" has been shipping for over a year, and it does not involve sharing a password.

Web Bot Auth builds on HTTP Message Signatures, the IETF standard published as RFC 9421. An agent signs each request with a private key and publishes the matching public keys in a directory served over HTTPS at a well-known path. The request carries a signature, a key identifier and a validity window, so a site can verify which operator sent it and that the request was not tampered with or replayed. Cloudflare validates these signatures at the edge and lets enterprise customers act on signed agents as a group, separate from search crawlers and training bots, as it described when it opened the program.

Payments have gone further. Visa's Trusted Agent Protocol and Mastercard's Agent Pay both sit on top of Web Bot Auth, adding a signed intent tag that separates browsing from paying and a nonce to block replay, so a merchant can tell "this agent is shopping" from "this agent is authorized to complete this purchase," per Cloudflare's technical writeup.

None of this is exotic. But it is still drafts and vendor programs, not regulation, and no US airline has published an agent authorization program. United's policy prohibits "unauthorized" agents, which quietly implies that an authorized category exists. As of this week, the carrier has not said how an agent becomes authorized.

A dark corridor opening into a lit one, standing for closed and open agent access

What It Means for Builders

If you are shipping agents that touch other people's systems, five things follow from this week.

Credential sharing is now a dead integration strategy. An agent that logs in as the user is, at a growing list of institutions, a policy violation and a liability transfer to the customer. Treat browser automation behind a password as a stopgap with a visible expiry date.

Build on the sanctioned lane first. Ask every vendor a single question: who signs for me? If the answer is "nobody, we just use the customer's session," assume the integration has a shelf life measured in quarters. Merchant-hosted MCP servers, partner APIs and documented agent programs are the durable path.

Sign your requests even before you are allowlisted. Publishing a key directory and sending RFC 9421 signatures costs an afternoon of engineering and moves you from "anonymous bot" to "verifiable operator" the moment a site turns verification on.

Keep a human in the loop where a human act is legally required. Baggage attestation, customs declarations, medical consent, and identity checks are not UX friction to be automated away. They are the compliance surface, and they are the reason United can plausibly argue an agent cannot complete check-in at all.

Instrument liability. Capture consent, cap what an agent may spend or change, and log every action per user. In banking the default rule is that the customer eats the loss, and your logs are the only thing that makes that argument survivable.

Three frosted glass panels on a dark desk, standing for sanctioned agent access lanes

The Next Twelve Months

The fight is no longer about whether agents will transact on the open web. It is about who issues their identity, and whether every airline, bank and marketplace must integrate a separate registry or converge on one.

Watch for three signals: the first airline or bank that joins a signed-agent directory, the first aggregator that exposes an agent policy to its supply side, and whether Web Bot Auth graduates from Internet-Draft to RFC. Until then, expect more quiet policy updates and more agents hitting walls that look, from the builder's side, exactly like a login form that suddenly stopped working.

Share𝕏

The Automation Brief

Read 5 AI stories instead of 50.

The essential moves in AI agents, models, automation and infrastructure — filtered for builders and operators, with the part that actually matters.

No noise. Unsubscribe anytime.

Editorial notes

Reported by

Stefan Trbojevic

Edited by

n8n Lab Editorial

Published

12 November 2026

Updated

12 November 2026

AI disclosure: AI assisted with research and drafting. Factual claims are reviewed by an editor.

n8n Lab is an independent service provider. We are not affiliated with, endorsed by, or sponsored by n8n GmbH. “n8n” is a trademark of n8n GmbH and is used here only to describe the platform-specific implementation and automation services we provide.