Share with your CIO
The open web is quietly closing for AI agents. Amazon has blocked Meta’s Muse, OpenAI’s Dots, and xAI’s Grok from shopping on its platform, while Delta is refusing agent-based ticket purchases, JPMorgan Chase and Charles Schwab are locking out agents on security grounds, and Yelp now demands a paid licensing agreement before granting access. The pattern spans retail, travel, and financial services. Meta has responded by cutting partnership deals with Walmart, Gap, and Sephora, while Shopify has opened its storefront network to agents, sketching out what a permission-based agent economy might actually look like.
What this means for your business
Every agent deployment your organization is planning rests on an assumption that the web will cooperate. It won’t, not uniformly, and probably not where it matters most. Financial institutions and large e-commerce platforms are precisely the sites where agentic workflows generate the most value, and they’re also the first movers on blocking. If your automation roadmap treats open web access as a given, you’re pricing in an asset you don’t own.
The more revealing dynamic here isn’t the blocks themselves but why they’re happening. Amazon’s objection to Perplexity was transparently about ad impressions, not security. Yelp’s data licensing demand is a monetization play dressed up as policy. These aren’t principled governance decisions; they’re incumbents discovering that AI agents route around the attention-capture mechanics that their ad revenue depends on. A user whose agent buys the flight, checks the restaurant, and clears the inbox never sees a single display ad. The platforms that built their businesses on that eyeball traffic have every rational incentive to treat agents as adversaries.
What emerges from this is a two-tier web. One tier is permissioned, governed by partnership deals like Meta’s retailer network and Shopify’s open catalog, where agents work because the platform decided they should. The other tier is locked, either explicitly or through anti-bot infrastructure that can’t distinguish a legitimate agent from a scraper. Organizations deploying agents inside the second tier won’t just hit friction; they’ll face account suspensions, legal exposure on the Perplexity model, and reliability failures that undermine the use case entirely. The vendor who can map which tier your target workflows live in is worth more right now than any benchmark score.
Watch the appeals court thread in the Amazon-Perplexity case as the leading indicator. The August reversal of the initial injunction suggests courts haven’t settled whether a user’s agent has the same access rights as the user, and that question is live. If it resolves in favor of agents, blocking becomes legally precarious and the permission-deal model Meta is pursuing looks like unnecessary capitulation. If courts affirm site owner authority, every agent strategy gets rewritten around negotiated access, and the CIOs who built informal partnerships with platform vendors early will find themselves holding a durable advantage over those who assumed the web stays open.
Based on reporting from Websites are closing their doors to AI agents, originally published 2026-10-17 09:00:00.
