United Airlines Restricts AI Agents From Booking Flights Over Security and Compliance Risks

AI agents are moving beyond conversation and beginning to perform transactions on behalf of users. That shift is now creating friction with services that handle payments, identity information and regulated processes. United Airlines says it will restrict unauthorized third-party AI agents and assistants from booking flights or managing reservations for customers.

The policy is aimed not only at United’s own digital platforms. The airline said the restrictions will also work through travel agencies and aggregators, alongside updates to its booking, ticketing and distribution policies. In effect, United is trying to prevent personal agents from accessing its systems through a customer’s shared login or other unauthorized paths.

Key points

• Credential sharing creates a new attack surface. If users give an AI assistant their password or account access, a leak could expose sensitive information and allow actions beyond what the user intended.
• The consequences can be transactional. United warned that third-party assistants could trigger unintended purchases, alter account details or use MileagePlus miles.
• Some airline procedures depend on human confirmation. The airline pointed to federal aviation requirements under which a person must complete certain check-in acknowledgments, including confirming that baggage does not contain prohibited substances.
• The concern extends beyond airlines. Banks, including Bank of America, have also warned customers against sharing passwords with AI assistants. Both sectors are struggling to distinguish a genuine customer from software acting on that customer’s behalf.

Why the decision matters

Traditional online services generally treat a successful login as evidence that the person operating the account is authorized. AI agents complicate that assumption. A user may give an agent broad access, but the service still needs to determine whether each action was properly authorized, understood by the customer and compliant with applicable rules.

For travelers, an agent could make flight searches and form filling easier. Yet airline bookings combine payment details, personal data, loyalty balances and time-sensitive travel plans. A mistaken purchase or account change may be difficult to reverse. For airlines, opening systems to general-purpose agents could also increase exposure to fraud, credential theft, repeated requests and disputes over who approved an action.

United’s position does not necessarily reject automation altogether. It suggests that companies may want stronger controls before allowing agents to transact: verified identity, explicit permissions, audit trails and human confirmation for high-impact steps. Until those mechanisms are widely available, blocking unauthorized agents is a simpler defensive choice than accepting unclear liability.

The larger issue is whether an AI agent can become a trusted digital representative rather than merely a tool operating with borrowed credentials. That will depend as much on authorization and accountability infrastructure as on the underlying model’s ability to complete a task. United’s restriction is an early sign that access policies, not just model performance, will shape the next phase of agent adoption.

Source: TechCrunch AI