Why AI Shopping Agents Keep Getting Stopped by Amazon, Banks and Bot Defences

The issue matters to any business deciding how to handle the growing volume of automated traffic that arrives on behalf of real customers. It also matters to the AI companies whose agent products depend on access to other firms' websites.

Deliberate Blocks

Amazon has deliberately stopped Meta's Muse assistant from shopping on its site, according to TechCrunch as cited by Engadget. The retailer is also fighting Perplexity over its Comet AI browser. Amazon argues that the agent breaks its access rules and puts customer data at risk, while Perplexity says users have authorised it to act for them. A federal appeals court vacated a preliminary injunction against Perplexity in August 2026, but the underlying dispute has not been resolved. Amazon itself offers a Buy for Me feature that can buy products from other brands' websites.

Other companies have written restrictions into their terms. eBay's user agreement prohibits unauthorised buy-for-me agents unless the company grants permission. Delta told TechCrunch it has no integration that lets third-party agents book flights on its site, but did not say whether it actively blocks particular agents.

Banks are concerned about security. On JPMorgan Chase's October earnings call, chief financial officer Jeremy Barnum said some agents rely on shared login credentials and screen scraping to transfer funds, methods he called "not secure." The bank blocks them where it can. Barnum also warned that customers who hand their credentials to an agent could be held responsible if it takes an action they did not want.

Accidental Blocks

Not every failure is intentional. Walmart, one of Muse's retail partners, said its human-verification checks were not designed to stop the assistant but still interrupted it. Zillow has acknowledged that its anti-scraping defences sometimes catch agents it otherwise allows. Google notes that traffic resembling automated activity, such as that from shared VPN networks, can trigger reCAPTCHA challenges, and agents navigating a site can meet the same checks.

Infrastructure providers are starting to separate these categories. Cloudflare now lets site owners treat search crawlers, AI training bots and agents acting for users differently. Since September 15, new domains on Cloudflare have by default allowed search crawlers while blocking AI training bots and agents on pages that display ads, although owners can change those settings. Shopify has taken a managed approach: its checkout tools let compatible browser agents fill in checkout details and submit orders after the buyer confirms, and hand control back to the shopper for required authentication steps.

What It Means for Businesses

The picture is uneven. Large platforms are using access controls to protect their own customer relationships and agent products, banks are drawing a firm line on credential sharing, and many sites still rely on bot defences that were not built with legitimate agents in mind. For companies building or deploying agents, the reliable route is likely to be formal integrations and permissioned checkout flows rather than agents that simply browse like a human. Retailers and service providers will have to decide which agents they let in and on what terms.