United Airlines AI Agents Ban: 6 Lessons for Builders

United Airlines AI agents policy just drew a hard line in the sand: the carrier says it will block third-party personal AI assistants from booking flights or managing reservations for its customers. It announced the move on Wednesday, November 11, 2026, and says it’s the first US airline to state this explicitly.

For anyone building agentic AI products, this is a big moment. The whole pitch of AI agents is that they act on your behalf, across websites and apps you already use. United just said: not on our systems, at least not yet. And it probably won’t be the last company to say so.

What United actually announced

According to Skift, United will restrict unauthorized third-party personal AI agents on its own digital platforms. It will also block them indirectly through online travel agencies and aggregators. The airline says this includes updates to its booking, ticketing and distribution policies.

United gave three main reasons. First, sharing account credentials with personal AI assistants “can expose sensitive information.” Second, these tools could trigger unintended purchases, account changes or MileagePlus redemptions. Third, automated actions could clash with FAA and Department of Transportation rules. The example it used is check-in, where a human must confirm their bags don’t contain banned items like explosives.

United says it will keep evaluating the policy as AI assistants evolve, and it plans to update its Digital Trust Center to explain the restrictions.

The business angle nobody should ignore

Let me be direct: security isn’t the only motive here. United CEO Scott Kirby said at a Chicago event, as reported by The Wall Street Journal, “I don’t want to create another middleman.”

That line explains a lot. Airlines spent two decades trying to pull customers away from online travel agencies and toward their own apps, where they control pricing, upsells and loyalty. An AI agent that compares fares across every airline and books the cheapest option looks a lot like a new OTA. Except this one lives inside your phone’s assistant.

So yeah, United’s decision is part cybersecurity, part compliance and part channel strategy. Builders should read it with all three in mind.

Where the rest of the industry stands on AI agents

The travel industry is split. Meta CEO Mark Zuckerberg has called travel a “top use case” for Meta’s Muse assistant, and TechCrunch reported on November 12 that Muse can now place calls to US businesses for paying users. OpenAI bought AI travel agent startup Odessia on November 10. Southwest Airlines launched a ChatGPT plug-in for flight search. Delta CEO Ed Bastian, meanwhile, says he’s “cautious” about agents in booking.

Outside travel, banks are worried too. TechCrunch noted that Bank of America recently urged customers not to share banking passwords with AI assistants. The common thread is identity. When an agent logs in with your password, the company on the other end can’t tell you apart from the bot, or from an attacker who stole the bot’s access.

6 lessons for teams building agentic AI

Here is the thing: blocking AI agents is a reaction to how most of them work today, not to the idea itself. If you build agents, these are the lessons worth taking seriously.

1. Stop relying on shared passwords. Screen-scraping with a user’s credentials is the fastest way to get blocked. Push for delegated access through OAuth-style flows, scoped tokens and official APIs wherever they exist.

2. Design for least privilege. An agent that searches flights doesn’t need permission to redeem miles. Separate read actions from spend actions, and make each one explicit.

3. Keep a human in the loop for money and compliance. United’s check-in example shows that some steps legally require a person. Build confirmation screens for purchases, cancellations and anything regulated.

4. Make every action auditable. Log what the agent did, when, and on whose instruction. If a customer disputes a booking, you need a clear trail.

5. Partner instead of scrape. Southwest’s ChatGPT plug-in shows that companies will work with agents on their own terms. Formal partnerships last longer than workarounds.

6. Expect policy whiplash. Terms of service will keep changing as companies figure this out. Build agents that degrade gracefully when a site blocks automation, rather than failing silently.

What this means for businesses deploying agents internally

Not every company is building consumer agents. Plenty are rolling out AI assistants for employees to handle travel, procurement, expense reports and vendor portals. United’s move is a warning for them too.

If your internal agent books travel by logging into airline accounts with stored employee credentials, it could stop working overnight. Worse, it could get the account flagged. The safer approach is to route agent activity through approved corporate travel platforms and official integrations, with clear spending limits and approvals.

Security teams should also check their own policies. Do employees paste work credentials into personal AI assistants? Bank of America’s warning to consumers applies just as much to corporate logins. Frameworks like the NIST AI Risk Management Framework are a useful starting point for setting rules.

Will other companies follow United?

Probably, at least in sectors with heavy regulation or strong loyalty programs. Airlines, banks, insurers and healthcare providers all have reasons to be careful. But outright bans are hard to enforce at scale, and customer demand for agents is growing fast.

The more likely long-term outcome is a two-tier system. Approved agents get official access through APIs, with proper identity checks and limits. Unapproved agents get blocked. That’s roughly how the web handled search crawlers and price scrapers in earlier decades. Agent developers who get ahead of that shift will be the ones still standing.

A note for AI startups outside the US

This isn’t only an American story. Plenty of agent startups, including teams in Pakistan, India and Eastern Europe, build travel, shopping and booking assistants for US customers. Many of them rely on browser automation because it’s quick to ship. United’s policy shows how fragile that model is.

If your product depends on logging into someone else’s platform, treat that dependency like a supplier risk. Track the terms of service of every site your agent touches. Have a fallback, such as handing control back to the user with a pre-filled booking link. And talk to your customers openly about what the agent can and can’t do. Trust is the real product in agentic AI. Lose it once with an unwanted purchase, and you rarely get it back.

Key Takeaways

• United’s policy: The airline is blocking unauthorized personal AI agents from booking or managing reservations, directly and through OTAs.
• Three stated reasons: Credential exposure, unintended purchases or miles redemptions, and conflicts with FAA and DOT rules.
• A business motive too: CEO Scott Kirby doesn’t want “another middleman” between United and its customers.
• Industry is split: Meta, OpenAI and Southwest are leaning in, while Delta and banks are cautious.
• Builder lesson: Move from shared passwords to scoped, auditable, API-based access with human approval for spending.

How TecniForge Can Help

At TecniForge, we help businesses navigate these technology shifts. Whether you need custom software development, AI integration, or cloud migration, our team builds scalable solutions. Talk to our experts.

If a major partner blocked your AI agent tomorrow, would your product still work?