AI agents are hitting a new hurdle: websites blocking them

Personal AI agents — Meta’s Muse, ChatGPT’s Dots, and others — promise to handle tasks like booking flights, making dinner reservations, and ordering groceries without the user doing the technical work themselves. But consumers are increasingly running into websites that block the agent before it can finish the job. Amazon recently began blocking Muse from its retail site, preventing the agent from browsing or completing purchases on Amazon’s catalog.

The problem extends well beyond Amazon and Muse: users across social media report agents hitting similar walls, some by design and some as a side effect of traditional anti-bot defenses meant to stop spam and abuse. A handful of complaints point to Muse failing to complete purchases on Walmart’s site — a pattern NBC also reported — even though Walmart is actually partnered with Muse, having announced the tie-up at Meta’s Connect developer conference in September. A Walmart spokesperson said the blocks are unintentional: when a human-verification button meant to prove a visitor isn’t a bot interrupts the agent’s flow, the check can fail and boot the agent.

To address the confusion, industry partners including Meta, Walmart, Stripe, Sierra, Genesys, Rocket, NiCE, and Decagon have begun work on an open standard for how AI agents communicate with businesses, aimed at separating legitimate agents acting on a user’s behalf from bad actors — specifically for agent-to-agent commerce.

Other companies are more cautious. Delta said it currently has no partnership letting a third-party agent book its flights, noting any future access would need to balance new technology with security and the customer experience. United pointed to its Terms of Use, which bars automated tools from accessing its site without permission. Yelp doesn’t allow non-human traffic unless an agent has paid for data-licensed access, and eBay restricts unauthorized agents and scraping while allowing some agent purchases. Adidas, Zillow, and Pizza Hut didn’t comment on similar complaints; some users reported eBay account suspensions over agent use, and one said Google’s Instinct was kicked out mid-task while cleaning a Gmail inbox.

Cloudflare, which protects sites from unwanted bot traffic and runs a marketplace where AI bots pay for the data they access, is suspected by some of disrupting Muse traffic after a September 15 change to its crawler defaults let site owners block AI agents on ad-supported pages; the company said it had no specific data to share on the pattern. Meta has argued that forming direct brand partnerships and pushing new standards, rather than leaving the blocking patchwork to sort itself out, is the better path: “Turning away a personal agent means turning away the customer behind it,” the company said in its announcement. “We’d rather work with businesses to address the underlying concerns than see them lose that customer.”

Why it matters

The pattern suggests today’s bot-verification systems, built to stop spam, may not fit an agent-first web, with Meta’s push for brand partnerships and a new agent-to-business standard emerging as one proposed fix.