United Airlines Blocks Customers' AI Agents: Should Your B2B Portal Allow Them or Refuse?
November 13, 2026 · 5 min read · Etzal Group Team
On 12 November 2026 TechCrunch reported that United Airlines will restrict access for third-party AI agents that book flights and manage reservations on behalf of users, citing cybersecurity and compliance risk. On the same day, TechCrunch also reported that Meta's AI agent Muse can now phone businesses in the United States on behalf of paying users.
Both events took place in the United States, and our sources report nothing similar in Indonesia or Singapore. The question, though, applies to anyone running a customer portal, a dealer portal or a booking application: when the one logging in is no longer the customer but software acting for them, what is your position?
What United said
According to the TechCrunch report:
- The restriction covers AI agents that book flights and manage reservations, both directly through United's digital platforms and indirectly through travel agents and aggregators.
- The aim is to "prohibit use by unauthorized, third-party personal AI agents or assistants".
- On security, United said "sharing account credentials with personal, third-party, AI assistants can expose sensitive information".
- On consequences, the actions of such tools "may result in unintended purchases, account changes, or use of MileagePlus miles".
- There is a regulatory reason too: federal aviation rules require a human to complete check-in by declaring that their baggage contains no prohibited items.
TechCrunch places this in a wider pattern. Airlines and banks are struggling to tell real customers from AI systems acting for them. The outlet refers to its own article of 27 October 2026 about banks, including Bank of America, asking customers not to share banking passwords with AI assistants.
The other side: agents that phone you
The second report points the opposite way. According to TechCrunch, Muse can now place calls to businesses in the United States on behalf of paying users, with subscriptions starting at $20 a month. Meta says Muse has more than 10 million users. TechCrunch also notes that many businesses and receptionists hang up once they detect that the caller is not human.
Two stories, one picture: AI agents arrive through whichever door is open, and the businesses on the receiving end have no shared rule yet for meeting them.
Why this concerns B2B portals
The key word in United's statement is "unauthorized". What is being refused is an agent that enters with the customer's password, without permission and without an identity of its own.
From a portal owner's side, that kind of agent creates three problems:
- You do not know who acted. In your logs the agent looks identical to the customer. If an order is wrong, there is no way to show who pressed the button.
- The permission is total. A password grants everything the account holds: changing delivery addresses, seeing special prices, cancelling orders. An agent asked only to check a status receives the keys to all of it.
- Credentials change hands. Your customer's password now sits with a third-party service you have never assessed.
Yet a B2B customer using an agent to reorder stock is trying to buy from you with less effort. Refusing without offering another route hands that convenience to another supplier.
Three positions you can take
On the AI agents customer portal question, the choice is not only allow or refuse.
| Position | What you do | Benefit | Cost |
|---|---|---|---|
| Prohibit | Terms of use forbid sharing passwords with agents, plus automation detection | Quick to apply, clear risk position | Customers who want automation have no official route |
| Ignore | No rule at all | Costs nothing today | Responsibility is unclear when something goes wrong |
| Provide an official route | API or delegated access with limited permissions per customer | Agents are identified, narrowly scoped and logged | Must be built and maintained |
This table is our judgement, not something from the sources. The second position is the riskiest, because it is the absence of a decision.
For many B2B portals, the sensible sequence is to prohibit password sharing now while building an official route for the handful of actions customers ask for most.
Designing the official route
If you choose to provide access, four things decide its quality:
- Separate identity. The agent signs in with its own key or token, issued by the customer from inside their account and revocable at any time. The password is never shared.
- Narrow permissions. Start read-only: order status, stock, invoices. Open write actions one at a time.
- Value and frequency limits. Orders placed through an agent are capped by value and by number per day. Above the cap, the order waits for approval.
- Records. Every action is logged with a marker showing that an agent performed it, and for which account.
Actions that require a human
United pointed to one action that the rules say a human must perform, the declaration at check-in. Every portal has its equivalents. Write your own list. Typical entries:
- Changes to bank account details or billing address.
- Changes to the delivery address of an order in progress.
- Orders above a set value or outside the usual pattern.
- Cancellations after goods have been processed.
- Declarations or consents with legal effect.
For anything on this list, an agent may prepare the action, but a human presses the final button, with an extra verification step. If your sector has declarations required by regulation, confirm with legal counsel who is permitted to make them.
What you can do this week
- Read your portal's terms of use. Is sharing credentials with third parties, including AI assistants, already covered?
- Review last month's logs. Are there sign-in patterns that look automated: odd hours, inhuman speed, many accounts from one origin?
- Ask your five largest customers whether they use, or plan to use, automation to place orders.
- Write the list of actions that require human confirmation.
- Decide your position, write it in one paragraph, and tell customers before they ask.
For the phone channel, give reception and customer service staff a standard response for non-human callers: what may be shared and what may not.
If you decide to build an official route for agents on your portal, Etzal Group's enterprise software and AI consulting teams can help design it.
Sources
- TechCrunch, 12 November 2026: report on United Airlines restricting AI agents
- TechCrunch, 12 November 2026: report on Muse placing calls to businesses
Frequently asked questions
Why is United Airlines restricting AI agents?
According to TechCrunch, it cites cybersecurity and compliance risk. United said sharing account credentials with third-party AI assistants can expose sensitive information and lead to unintended purchases or account changes.
Has this happened in Indonesia or Singapore?
The sources we used report events in the United States only. We present this as preparation, not as something that has occurred in this region.
What is the alternative to banning AI agents outright?
Offer official access through an API with limited permissions per customer, and define which actions still require confirmation by a human.
Related service
WhatsApp AI Chatbot
We build, train, and run a WhatsApp AI agent that answers customers 24/7, qualifies leads, and follows up.
