Industry works on standard to ease AI agent access to retail sites

Personal AI assistants such as Meta’s Muse, Instinct and OpenAI’s Dots are being used to perform tasks like booking flights, reserving tables and ordering groceries. Users, however, report that many websites block these agents, preventing transactions. Amazon recently stopped Muse from browsing its retail catalog, and similar obstacles have been noted on Walmart, Delta, United, Yelp, eBay and other platforms.

Walmart, a declared partner of Muse, said the issue often stems from human-verification buttons that interrupt the agent’s flow. Delta’s global communications manager, Heena Chavda, stated the airline is evaluating AI agents but must prioritize security and customer experience. United referenced its terms of use, which forbid unauthorized automated access without written permission.

In response to growing friction, Meta, Walmart, Stripe, Genesys, Rocket, NiCE, Decagon and others are drafting an open standard to define how AI agents may interact with businesses, aiming to separate legitimate “good bots” from malicious traffic. Meta’s blog post emphasizes the need for clear norms that give businesses predictable control while allowing agents to serve customers.

Cloudflare, a major content-delivery network, noted recent changes to its crawler defaults that let site owners block AI training bots, though the company said it lacks specific data on the current agent blocks. eBay clarified its policy permits authorized shopping agents but restricts unauthorized scraping and model training.

The industry’s effort to create a common protocol reflects the broader challenge of adapting existing anti-bot measures to an “agent-first” web environment.