United's 259-word release says it will restrict third-party AI agents and names no incident or method

259 words, seven paragraphs: restrict appears, block does not (derived)

United's news release of 11 November 2026, headed "United Restricts Access for Third-Party AI Agents", runs to seven paragraphs and 259 words (derived, counted on spaces in the saved page). The verb United uses is restrict: "restrict access" and "these new restrictions". The word block, which is the verb in The Record's headline, does not appear in the release at all (derived, checked in the saved copy). Nor do incident, breach, cyber, privacy, detect, API or effective.

What that does not establish. It does not establish that any customer has lost anything, that any agent has made a booking, that United can tell an agent from a person, or when any restriction starts. The release gives no incident, no count, no method, no start date and no list of the agents it means. It states a policy direction and the reasons for it in United's own words. It is also a press release from a company that sells the tickets, sets its own booking, ticketing and distribution policies and runs MileagePlus, so a rule about who may act on a customer's account is also a rule about who stands between the airline and the customer.

Reading and time stamps. This site read United's release at 20:02 UTC on 12 November 2026, from the link in The Record's report. The release carries the date line November 11, 2026 and a page comment time stamp of 13:00 US Eastern Standard Time that day, which is 18:00 UTC. The Record's article was published at 19:24 UTC on 12 November, 25 hours and 24 minutes later (derived). United's own page was served with a last-modified header of 16:55 UTC on 12 November, 22 hours and 55 minutes after the release time stamp (derived); this site cannot say whether the text changed, and The Record's quotations match the saved text apart from a capital letter at the start of one. United's Digital Trust Center page, which the release says will be updated, could not be reached from this site (two requests for the host's robots.txt failed, one with a stream error and one with a timeout, so the page itself was not requested), so nothing here says what that page contains.

What the release says, in its own order

The first two paragraphs are about United's own app. United says it designed the app "to securely empower customers to manage their own travel experience", from booking a trip and choosing seats to rescheduling a flight in bad weather. That is the sanctioned route, and the release gives no data on how often customers use it.

The third paragraph carries the risk. In United's words, "sharing account credentials with personal, third-party, AI assistants can expose sensitive information", and "actions taken by these tools may result in unintended purchases, account changes, or use of MileagePlus miles". It then adds that the automated actions "have the potential to violate FAA and DOT regulations", with one example: "flight check-in requires human acknowledgment that bags do not contain banned substances like explosives". The verbs are "can", "may" and "potential": every risk is stated as a possibility, none as something that has happened.

The fourth paragraph carries the action: 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". It adds "This work includes updates to our booking, ticketing and distribution policies." The two routes are different. Directly means inside United's own digital platforms. Indirectly means through travel agencies and aggregators, and the only instrument the release names for that second route is policy.

Zoom

The fifth paragraph says "Corporate leaders in the banking and retail industries have taken similar actions regarding AI agents" and encourages others to do the same, without naming any company. The sixth promises a Digital Trust Center update so that customers understand "what these new restrictions mean for them, how to protect their accounts, and where to get help if an agent takes an unintended action". The seventh says United "will continue to evaluate how these tools interact with our services". The release itself gives customers no instruction beyond what the risk paragraph implies.

Three routes United describes, and a fourth the release leaves open

Read as a map, the release describes three routes to a booking. A customer uses United's app. A customer hands a login to a personal, third-party assistant, which United says it will restrict directly. Or an assistant reaches United through an agency or aggregator, which United says it will restrict indirectly through policy. Those are the first three rows below.

Zoom

The fourth route is the one the wording makes unavoidable. The release prohibits use by "unauthorized, third-party personal AI agents or assistants". An unauthorized agent implies an authorised one, and the release does not say who authorises, how, or whether any such agent exists. It does not say whether an agent that works through an interface United itself publishes would count. The word API does not appear, and neither does any other technical term for how an agent reaches United.

The word "personal" also narrows the target on the page, since it describes "personal AI agents or assistants". Whether an agent run by a company, a corporate travel tool or an agency falls inside the restriction is not stated. The check-in example is about one step, check-in, and United calls it "an example": whether booking carries a comparable human acknowledgment is not stated.

Where The Record and the release differ

The Record's report is the reason this brief exists, and its direct quotations from the release match the saved text apart from one capital letter. The differences are in its own summary lines, which is ordinary headline compression, but a reader who stops at the headline gets a different verb and different reasons from United's page. Both sets of wording are below as printed.

The Bank of America statement is attributed here to The Record alone. This site did not read a Bank of America page, and it cannot check The Record's "last week". Nothing in United's release depends on it.

Friendly labels doing the work of a control

Three labels in the release stand where a control or a piece of evidence would normally be. Unauthorized is a word that needs a published test: who authorises, and by what mechanism. The release does not give one. Digital Trust Center is the name of a page that United says it will update. It is a promise to explain, and this site could not read the page, so it is not yet evidence of anything. Securely describes the app in the second paragraph, in the same sentence as "industry-leading, award-winning", with no data behind either.

This site's reading, not United's: a restriction written into "booking, ticketing and distribution policies" binds the parties who sign those policies, such as agencies and aggregators. It is a different thing from a technical control on a customer who types a login into an assistant on their own device. The release names the policy lever and is silent on any technical one, so whether the restriction can be seen working is not something the page lets a reader check.

A second company's view, and what it sells

The Record quotes the founder and chief executive of Trua, which it describes as "identity verification company Trua". His view, sent to the reporter by text, is "Blocking agents or forcing more data sharing just delays the inevitable", and The Record says he argues for wider use of verified identity tokens. Trua sells identity verification, so a remark that favours verified identity is also its commercial position. It is an opinion about the outcome, and it is not evidence about United's policy.

The Record also writes that "Many companies, particularly banks and airlines, have faced mounting difficulty as they try to separate human customers from their AI assistants." That sentence is The Record's characterisation. It carries no figure, and United's release does not say it.

What to do, in order

United's release does not print instructions for customers. It says a Digital Trust Center update will explain how to protect accounts, and The Record attributes the advice to never share user IDs and passwords with agents to Bank of America. The list below is this site's suggestion, built from the risks United names. It is not United's advice.

One alternative design is described in this site's earlier briefing on a vendor's own safety post, which says credentials and approvals stay outside the model: Anthropic says Claude Duet keeps credentials and approvals outside the model. That is a vendor's description of its own product, and nothing in United's release refers to it.

Technical details

United's release prints no configuration, rule, request pattern, user agent string, indicator or detection logic, and the page does not describe a technical control. There is nothing of that kind to copy, so none is invented here. What it does print is the sentence that carries the action and the page data that fix its date. A defender or a risk owner can use the first to quote the policy exactly in a conversation with a travel provider, and the second to say which version of the page was read.

As a result, 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. This work includes updates to our booking, ticketing and distribution policies.

The block below is page data as read: the title, the date line, a time stamp that the page prints in a comment, and two response headers. The last-modified header is 22 hours and 55 minutes after the release time stamp (derived). This site cannot say what changed, if anything. A reader who quotes the release should record the date they read it.

og:title: United Restricts Access for Third-Party AI Agents
page date line: November 11, 2026
page comment: ITEMDATE: 2026-11-11 13:00:00 EST
HTTP status: 200
HTTP last-modified: Thu, 12 Nov 2026 16:55:27 GMT
HTTP date of this site's request: Thu, 12 Nov 2026 20:07:46 GMT

The question the release leaves open

United will restrict "unauthorized" agents. If the word means anything, an authorised agent exists, so who decides, by what test, and where is that published for the customer whose assistant is refused halfway through a booking?

Key facts