The next hurdle for AI agents: getting websites to let them in

Personal AI assistants — such as Meta’s Muse, Instinct, and ChatGPT’s Dots — are being positioned to act on users’ behalf for tasks like booking flights, making reservations and shopping. While these agents can automate interactions that previously required manual input, users have reported frequent failures when agents try to interact with third-party websites. Some of those failures are deliberate blocks by sites, while others appear to stem from conventional anti-bot systems that cannot distinguish legitimate agent activity from malicious automation. High-profile examples have surfaced publicly: Amazon recently blocked Muse from accessing its retail catalog and completing purchases. Social-media complaints also described Muse failing to complete purchases on Walmart’s site, though Walmart says those issues are not intentional and that it is partnered with Muse. In many reported cases the failure follows an interrupted human-verification flow (a click or similar interaction), which causes the site’s bot-protection check to fail and eject the agent. Beyond retail, travel sites and airlines have been cited as places where agents run into resistance. Delta told reporters it does not currently support third-party AI agents booking flights on its platforms and is weighing any opening carefully with security and customer experience in mind. United’s published terms prohibit automated access without written permission. Other platforms such as Yelp require paid data-access agreements for non-human traffic, and eBay says it restricts unauthorized automated actions while not banning all shopping agents outright. To reduce user confusion and enable legitimate commerce via agents, a group of industry partners including Meta, Walmart, Stripe and others are developing an open protocol focused on agent-to-agent communication for online commerce. The stated goal is to create a way for businesses and agents to identify and authorize “good” agents acting for users while continuing to block abusive or data-scraping automation. Meanwhile, users remain caught between service providers and merchants when a site’s existing anti-bot tooling blocks agent activity, producing inconsistent experiences across different services.