Airlines, Banks, and Retailers Turn Away AI Agents: The Web Is Closing Its Gates

Posted on Oct 19 • Originally published at joho-todai.com • AI-assisted

AI agents promised to handle everything on the web, from booking flights and restaurant tables to buying shoes. The web itself, however, is not accepting that promise.

Restaurant booking service Yelp accepts only AI agents with paid data licensing agreements. Agents without contracts fail when attempting reservations or waitlist registration.

Delta Air Lines currently does not allow third-party AI agents to purchase or book tickets on customers' behalf. Spokesperson Heena Chavda says Delta is evaluating how external agents might improve the customer experience. United Airlines pointed to terms prohibiting robots, spiders, and other automated devices, declining to address individual AI agents.

Financial institutions are also closing access. JPMorgan Chase and Charles Schwab reportedly block AI agents from accessing accounts over security concerns.

Social media reports describe blocks at eBay, real estate site Zillow, Pizza Hut, Adidas, and multiple airlines. Some users report eBay account suspensions for agent use. Others say Cars.com and US rail booking service Amtrak, previously accessible through agents, no longer work.

Walmart announced an official partnership with Muse at Meta Connect in September 2026. Nevertheless, users attempting to shop there through Muse reported blocks. The cause was a human-verification button installed to stop spam and bots. Agents failed the check and were rejected. A Walmart spokesperson described this as unintentional.

Zillow faced a similar structure. In a response after publication, it said agents working for consumers were permitted, but anti-bot measures against scraping had inadvertently caught some agents.

That is what makes the problem difficult. Users cannot distinguish a corporate decision to block agents from a side effect of existing security measures. Both produce the same result: the agent does not work.

One factor expanding the scale of blocking is Cloudflare's September 15, 2026 default change (in Japanese).

Cloudflare began blocking mixed crawlers, which serve both search indexing and AI training, by default. On sites already blocking AI bots for security reasons, the change meant AI agents were also blocked on pages carrying advertising. Site operators had not specifically chosen to block agents; Cloudflare's configuration change narrowed their reach in practice.

Joho no Todai confirmed that Cloudflare itself is testing a dedicated web browser for AI agents and operates a paid marketplace for AI bot data access. It stands on both sides: blocking agents and opening paths for them.

Legal disputes are already underway. Amazon sued Perplexity over its AI shopping agent Comet in November 2025. Comet logs into Amazon with users' permission and shops for them, but Amazon alleged a violation of the federal Computer Fraud and Abuse Act (CFAA). A district court issued a preliminary injunction in March 2026, but the Ninth Circuit reversed it in August, reasoning that the users themselves accessed Amazon's systems, rather than Perplexity. The underlying lawsuit continues, and the appellate court explicitly limited its ruling, saying it did not establish general legal principles for AI agents.

Amazon also blocked Muse (in Japanese) from browsing and purchasing in September 2026, just 12 days after launch. It sued Perplexity under the CFAA and cited terms-of-service violations against Muse. Through litigation and terms alike, Amazon continues excluding agents. Perplexity countered that Amazon wants agents out because they reduce the number of human eyes on ads.

Figure omitted — see the original article

Other companies have chosen to welcome them. Meta has begun developing open standards with Walmart, Stripe, Sierra, and others for safe interactions between AI agents and business systems.

Turning agents away means turning away the customers behind them. Rather than watching businesses lose customers, we want to work together on the underlying concerns

Meta offered that explanation when announcing the open standards.

AI agents were designed to use the web for people. The web, however, was designed on the assumption that people would visit. Advertising depends on human attention. CAPTCHAs distinguish people from bots. Terms of service are written as contracts humans accept.

Agents fit none of those assumptions. Despite representing people, they are indistinguishable from bots to the web's mechanisms.

eBay says it does not prohibit every third-party agent, instead restricting unauthorized agents, automated scraping, and model training. The question is shifting from whether something is an agent to whether it is authorized.

Users of Muse, Dots, and Instinct repeatedly encounter barriers in everyday shopping and booking because no mechanism yet exists for deciding who grants that authorization and how. Services with partnerships work; those without them stop. Whether the web changes from one built for people into one agents can also navigate depends on whether these standards arrive in time.

Originally published on Joho no Todai. The Japanese original is here.