Amazon, United, Target and others are telling AI shopping agents to stay out

A wave of 'no agents' rules

As businessinsider.com reported, a growing group of companies is putting limits on AI agents. An agent is a program that takes actions for you, such as searching, filling a cart or making a reservation, instead of only answering questions. Some of these companies stop agents from buying things for you. Others stop them from browsing the site at all.

The pushback comes as agents become easier to find. The article points to Meta's Muse, released in September, and similar assistants like Instinct. It also notes that popular chatbots such as ChatGPT and Claude have added agent features of their own.

The reasons companies give vary. Privacy and security come up most. There is also speculation that if people stop visiting sites to browse, buy or book, those sites could lose traffic.

Who is blocking what

The rules differ from company to company, so the list is worth reading closely.

• Amazon: GeekWire first reported in September that people using Muse to shop saw a popup saying "continued access by an unauthorized AI agent violates Amazon's Conditions of Use." Amazon's written terms do not ban every agent. They say the company may limit how any agent uses its services, including by technical measures.
• Booking.com: Automated assistants may not make bookings or reservations without express permission, and may not access the platform at all without prior permission. It says it will block anyone it suspects of using one.
• eBay: Its updated user agreement prohibits automated access without permission. It names "buy-for-me agents," bots driven by large language models (the technology behind chatbots), and any flow that places orders without a human reviewing them.
• Yelp: It does not allow non-human traffic without express permission. It runs a paid data-licensing program that can let an agent in through approved connections, including one that lets ChatGPT users make reservations.

A guest list, not a locked door

Not every company says no to everyone. Target prohibits AI shopping agents from accessing or collecting data from its site unless they are explicitly authorized. Its approved list names Target's own AI Shopping Assistant, plus OAI, Google and Microsoft. With a customer's permission, approved agents can sign in, build carts, place orders and start returns. Target also reserves the right to change the list or restrict agents at any time.

Resy, the restaurant booking platform, takes a similar approach. It says it supports discovery and booking through ChatGPT and Claude, but can block AI activity from other sources. It cited platform security and concerns about excessive demand on its systems. It also warned that users who break its rules could have their accounts terminated.

The pattern is that companies want to choose which agents get in, often ones they have a relationship with, instead of letting any tool act as a customer.

Why United draws a firm line

United Airlines announced its restrictions on Wednesday. It limits access for third-party agents and cites security and regulatory risks, as well as the chance of unintended purchases.

The airline gave a concrete example. If an agent checked in for a traveler, it could run afoul of federal rules that require a human to confirm that their bags hold no banned items, such as explosives. United's new policy says unauthorized AI agents may not search, book, buy or manage trips on its platforms.

The key word is unauthorized. The policy is aimed at agents the airline has not approved, not at every kind of AI.

What it means for everyday shopping

Think of a shop that welcomes you through the door but will not accept a stranger carrying your wallet. You can still shop, but a program acting as you may be turned away, and the rules can change without notice.

In practice, an AI helper may find the right product or fare and then stall at checkout. It may also work on one site and fail on the next. Which agent you use can matter as much as what you ask it to do.

Some friction is probably deliberate. Companies are trying to decide who is allowed to act as a customer, and what they should be paid or credited for letting an agent use their data. The Yelp licensing program shows one path: pay for approved access.