Delta, Yelp, JPMorgan and Charles Schwab block AI agents sent by users

The list of websites that will not let an AI agent act on behalf of a user has grown to include Delta Air Lines, Yelp, Charles Schwab, JPMorgan Chase, Cars.com and Amtrak. The Verge reports that users of Meta's Muse, xAI's Grok, OpenAI's Dots and Instinct have all run into blocks in recent weeks, and that each of those sites has either confirmed the block or said it allows agents only through a paid data licence.

Yelp told TechCrunch that an AI agent can reach its data only through its licensing programme. Delta blocks agents from booking tickets. JPMorgan Chase and Charles Schwab both say they block agents from account access for security reasons. The Verge adds that users have found Cars.com and Amtrak quietly stopped working with their agents too, which it reads as a signal that other sites are turning agents away without saying so.

Amazon is the longest-running example. It banned Perplexity's shopping agent and sued the company. A judge told Perplexity to stop its agents on Amazon, an appeals court overturned that ruling, and Amazon has since filed an amended complaint. Meta's Muse was blocked the same way when Amazon said its "continued access" violates site policies, and users have reported the same problem for Grok, Instinct and OpenAI's Dots.

Meta's response has been to sign with other retailers instead of fighting Amazon: Muse can now shop on Walmart, Gap, Sephora and Best Buy, and Shopify has opened its storefront network to agents. The Verge notes a second kind of block that is harder to see. Agents set off the anti-bot checks that sit in front of almost every site, so a Meta partner like Walmart still stops some Muse sessions at checkout, and Google has blocked Muse accounts for "suspicious activity".

Chris Canal, who runs the AI safety firm EquiStamp, told The Verge that this directly cuts the value of the lab building the agent. The claim from those labs, he said, is that the agent will "permeate every aspect of our lives, and if there are companies that are actively working to make that not happen, there are just fewer useful things the agents can do". Cloudflare is on the sites' side of this: it already offers a one-click block for AI crawlers, and has now started blocking crawlers that scrape for both search indexing and AI training. Canal expects "prove you are not a robot" checks to spread to ordinary consumer sites.

The reader building a web-touching agent now has two separate problems. The first is which top sites will let the agent in at all. That list is written by the sites, so a Meta partner today may be a Meta non-partner next quarter. The second is that even an allowed site may stop the agent at an anti-bot check, so a partnership announced in a press release can still be an unreliable integration in production. Both lists change week to week, so the honest answer in a demo today is which sites the agent was tested against this week.

Source

The Verge, by Emma Roth.