Businesses that block third-party AI agents risk losing access to an emerging sales and product discovery channel, while those that let agents operate on their platforms must contend with security risks, operational challenges and added costs.
United Airlines brought that tradeoff into focus on Nov. 11 when it updated its website terms and contract of carriage to restrict unauthorized third-party AI agents from accessing its platforms or handling tickets. It cited risks to customer data, unintended purchases, unauthorized account changes and violations of federal aviation regulations, as well as agents' inability to attest that passengers' bags contain no prohibited items during check-in.
The revised contract bars unauthorized agents from making, changing or servicing tickets, even under human supervision, and allows United to cancel bookings they make. It also requires written permission before agents access its website or collect information, but traditional search engines can still index its content.
Importantly, customer permission alone doesn’t make an AI agent authorized under United’s policy. The airline says it doesn’t currently work with any third-party AI agents, although its rules make exceptions for certain human-reviewed AI tools used by travel agency employees and rule-based booking automation.
The concerns also extend to distribution and customer relationships. United's CEO, Scott Kirby, told the Wall Street Journal that he didn't want to introduce another intermediary between the airline and its customers.
But other airlines are taking different approaches.
Delta supports certain authenticated interactions with external AI assistants as it develops its own agentic customer-service capabilities. American Airlines has indicated that some AI tools can facilitate ticket purchases, although it hasn’t specified which booking steps those tools can perform. Meanwhile, JetBlue said the airline should adapt if customers increasingly use AI booking agents, while raising concerns about distribution costs.
The airlines' differing approaches reflect a broader enterprise challenge: deciding whether to block external agents, give them controlled access or build proprietary tools.
What blocking agents could cost
Restricting agent access could make it harder for businesses to reach customers who use AI-driven product discovery. Personal AI agents let consumers research, compare and buy products using a single interface rather than visiting multiple websites and apps.
"AI agents are increasingly becoming the entry point for product discovery," said Jaclyn Wands, vice president of product and AI at Phaedon, a loyalty tech and services company. "When consumers delegate research to an assistant, the agent builds a shortlist before they visit a company's site or app."
Unlike lost website traffic, she said, businesses might not know they lost a customer. "Customers who encounter a block could turn to competitors or other platforms, giving the original business less control over how its products are presented."
The restrictions could also weaken relationships with existing customers. "Even longtime customers could look like first-time buyers if agents can't access their loyalty status, rewards and preferences," Wands said.
Access versus control
Rich Pleeth, CEO and co-founder of Finmile, a provider of AI workers for logistics and physical operations, argued that companies should give approved AI agents limited access to product information while retaining control over purchases and account changes. He said companies should distinguish between permission to compare products and permission to purchase them.
"Companies could let approved agents retrieve prices and availability while requiring explicit customer authorization for purchases, substitutions or account changes," he said.
The FAA's nationwide ground stop on Nov. 13 illustrates why that distinction matters. A communications outage affecting air traffic control centers halted departures for hours and disrupted more than 10,000 flights. United's contract bars unauthorized agents from servicing tickets, even when a person supervises them, raising questions about whether they could help passengers rebook during similar disruptions.
Reliable information is also important. "Agents might favor a competitor if a seller can't confirm product availability or delivery, even when it offers a lower price," Pleeth said.
Securing third-party AI agent access
But giving external agents access to customer and product information also creates identity and security challenges. Businesses need to verify which customers personal agents represent and what actions they can take on their behalf.
United says its systems can distinguish an agent booking from one made by a traveler or travel agency. Delta relies on login credentials and places the responsibility on the customer, while United warns that sharing credentials with an AI assistant can expose sensitive information.
Saurabh Pitkar, director of product management for agentic commerce and agentic AI at Dell Technologies, said account linking can identify customers and give agents access to information such as loyalty benefits without exposing their entire accounts.
"Businesses will need to figure out how to bring guardrails and protocols that allow external third-party agents to securely perform tasks on behalf of customers," Pitkar said. He cited role-based access controls, security measures and fraud controls as important safeguards.
While those safeguards address security, businesses must also weigh the operational and financial implications of agentic commerce, including whether to build their own agents to interact with customers' personal agents, integrate with approved third-party agents or restrict external access, and how much each approach will cost.
The cost of acquiring customers using AI agents
Building their own agents is one option for businesses, but Wands cautioned that customers might stick with third-party assistants they already use. Partnering with external AI platforms could help businesses reach those customers, but it could also mean paying distribution fees and becoming dependent on outside platforms.
Pleeth warned that companies could end up paying fees for customers they already have. "Someone who previously booked directly on an airline's website might start booking through an assistant, creating a new fee without bringing in a new customer," he said.
For example, if the provider of a personal AI agent charges a referral or transaction fee for bookings made through its service, an airline could incur an additional cost for a customer who might otherwise have booked directly through its website. The airline gains no new customer but pays a fee for the booking.
A PYMTS survey of 60 senior e-commerce decision makers at U.S. merchants found that most expect merchants to pay fees or commissions on completed agent-driven sales.
"Businesses should measure completed purchases, repeat transactions and margins after platform fees, fulfillment, returns and support costs," Pleeth said. He also recommended tracking whether agent-driven sales increase overall revenue or simply shift purchases from direct channels to platforms that charge a fee.
Kinza Yasar covers AI and emerging technology for TechTarget, with a focus on ethics, enterprise adoption, governance and business strategy. Before moving into journalism, she worked in IT and network support roles, giving her a systems-level perspective on how enterprise technologies are built, deployed and managed.