United Airlines updated its Contract of Carriage on November 11, 2026, with a rule that lets the airline cancel any AI-booked reservation without notice. CEO Scott Kirby dropped the official framing quickly at a Chicago event: “To be honest, I don’t want to create another middleman.” The first major U.S. carrier has drawn a formal legal line around autonomous bookings — and the stated security rationale is only part of the story.
United Airlines AI Agent Ban: What Rule 6.J.6 Actually Says
Rule 6.J.6 of United’s updated Contract of Carriage prohibits AI agents from making, modifying, or servicing any ticket. Rule 5.H gives United the right to cancel such reservations without notice. The scope is broad: the ban covers United.com, the United app, travel agencies, and aggregators. If an AI-enabled browser or an assistant like Meta Muse facilitated your booking through Expedia, that reservation is still potentially voidable.
The most dangerous clause for consumers: the ban applies even when a human approves every step. It is not about whether a person reviewed and confirmed the booking — it is about whether an AI agent touched the transaction at any point. Developers building travel AI agents on United routes are not just operating in a gray area; they are building on explicitly prohibited ground.
Related: Railway Post-Mortem: AI Agents Stole $740k in API Keys
The Middleman Admission
Kirby’s “middleman” comment was not a slip. United spent years fighting online travel agencies like Expedia and Orbitz over distribution fees, fare display rules, and customer ownership. AI agents are, in United’s view, the next generation of that same threat. Meta stock dropped 1.97% on the day of the announcement — the market correctly read this as a direct shot at Meta Muse, whose highly promoted travel booking use case just lost a major carrier.
The economics explain the hostility. Airlines earn 60-70% of profits from ancillary revenues: baggage fees, seat upgrades, credit card promotions, lounge access upsells. An AI agent that finds the cheapest fare and completes the booking in two API calls bypasses every one of those upsell screens. United described this as AI agents “stripping out the ancillary margin while retaining full customer support liability.” That is an airline executive translating “this hurts our business” into compliance language.
The Legitimate Concerns (And Their Limits)
Not every concern United raised is cynical. Under 49 CFR 175.25, federal law requires passengers to personally attest during check-in that their bags contain no hazardous materials. An AI cannot make that attestation on behalf of a human — this is a genuine regulatory constraint. Credential security is also real: AI agents that store login credentials create risks if those services are breached, as the Railway incident demonstrated.
However, neither concern fully explains why United can cancel a booking that a human reviewed and approved. The FAA attestation issue applies to check-in, not to the ticket purchase itself. The credential risk would be addressed by proper OAuth delegation, not an outright ban. The gap between the stated rationale and the actual policy scope is exactly as large as the ancillary revenue at stake.
What Developers Building AI Agents Should Do Now
Three constraints apply going forward for anyone building travel AI agents. First, avoid credential delegation — agents that log in with user credentials trigger policy violations. Second, implement query budgets — one developer testing an agent pulled 881,076 fare combinations for a single itinerary, an infrastructure attack United will block at the network level. Third, require human confirmation gates before any transaction completes, especially for payments and loyalty redemptions.
The industry is not uniformly hostile. Delta permits AI assistants that authenticate customer credentials. American Airlines allows AI through certain authorized partners. Airbnb and OpenAI are building their own first-party agents, keeping transactions within their own ecosystems. The viable path is direct API partnerships with carriers, not browser automation. Browser-scraping agents are legally fragile on United routes and operationally fragile everywhere — airlines can change their UI without notice, breaking agents immediately.
Key Takeaways
- United Airlines’ Rule 6.J.6 bans AI agents from booking or modifying tickets; Rule 5.H lets United cancel AI-facilitated bookings without notice, even when humans approved each step
- CEO Scott Kirby’s own words reveal the real motive: AI agents are the new OTAs, and United is not interested in another intermediary capturing its customer relationships and ancillary revenue
- The FAA attestation requirement (49 CFR 175.25) is a genuine regulatory constraint on AI check-in, but it does not explain the full scope of the transaction ban
- Developers must avoid credential delegation, implement query budgets, and require human confirmation gates — browser automation on United routes is explicitly prohibited
- The travel industry is splitting: United blocks third-party agents; Delta offers authenticated access; Airbnb and OpenAI build their own. Direct API partnerships are the only defensible architecture
