OpenAI, Meta, Google and Instinct are all racing to sell you an agent that browses the web on your behalf, booking tables, buying shoes, clearing your inbox. On October 17, 2026, The Verge reported what happens when the rest of the internet refuses to play along: Yelp, Delta Air Lines, JPMorgan Chase and Charles Schwab are all blocking AI agents now, and users say sites that used to work with their agents, including Cars.com and Amtrak, have quietly stopped cooperating. The short version: as of October 17, 2026, agents from the biggest AI labs are being blocked or restricted by a growing list of major websites, even as the labs market them as your personal web surrogate.
The block map: who moved, and how
The Verge's October 17 report, the originator of this story, names the blockers and their mechanisms. The list, as reported:
| Company | What agents are blocked from | Stated reason |
|---|---|---|
| Yelp | All agent access except through its paid data licensing program | Agents must pay (told to TechCrunch) |
| Delta Air Lines | Booking tickets on behalf of users | Agents cannot transact as passengers |
| JPMorgan Chase | Accessing accounts via agents | Security concerns |
| Charles Schwab | Accessing accounts via agents | Security concerns |
| Amazon | Meta's Muse agent shopping ("continued access" violates site policies) | Policy violation; separate Perplexity lawsuit ongoing |
Below the named companies, a quieter signal: users of Meta's Muse, xAI's Grok, OpenAI's Dots and the Instinct agent report that Cars.com and Amtrak, which previously worked with their agents, no longer do. Nobody announced those blocks. They just happened, which suggests the published list is the floor, not the ceiling.
The precedent: Amazon vs Perplexity is still in court
This is not a new fight. Amazon banned Perplexity's agentic shopping feature in 2025 and sued the startup, saying it had repeatedly demanded Perplexity stop using AI agents to buy on its site. A court initially ordered Perplexity to stop; an appeals court overturned that order in August 2026; Amazon then filed an amended complaint. The case is live today.
The two sides stated the philosophical split that every other company is now picking from. Amazon, via The Verge: third-party companies buying on customers' behalf "should operate openly and respect service provider decisions whether or not to participate." Perplexity's counter: Amazon doesn't want agents on its platform because agents mean fewer human eyeballs on ads, and Amazon "does not believe in your right to hire labor, to have an assistant or an employee acting on your behalf." Is an agent your employee, or Amazon's trespasser? Courts have not finished answering.
Even without a ban, the bot detectors are waiting
Here is the evergreen part, the mechanism that makes all of this inevitable. A website cannot easily tell your helpful agent from a credential-stuffing bot, because technically they are the same thing: automated software driving a browser or API on someone's behalf. Detection layers stack up: user-agent strings (trivially faked), behavioral fingerprinting (mouse movements, typing cadence, request timing), rate limiting, CAPTCHAs and commercial anti-bot systems, robots.txt, login walls. When an agent books a flight, it trips the same alarms as a scalper bot, and the site's security team has no reliable way to distinguish them. That is why the banks moved first and why they cited security, not strategy. The blocklist is not a bug in the agentic web. It is the agentic web meeting the adversarial web.
The holdouts: Meta would rather negotiate
Not everyone is building walls. Meta's response earlier this month, as reported by The Verge, favors cooperation: turning away a personal agent means turning away its customer, and the company says it would rather address businesses' concerns than lose that customer. The labs keep shipping regardless: OpenAI's Dots, Meta's Muse, Google's agents and the Instinct agent all promise browsing on your behalf right now.
Chris Canal, CEO of the AI safety evaluator EquiStamp, put the commercial stakes plainly to The Verge: "Does this impact the perceived value of the Anthropics and the OpenAIs? I would say absolutely. One of the claims is that they're going to permeate every aspect of our lives and be able to do anything that we could do online. If there are companies that are actively working to make that not happen, there are just fewer useful things the agents can do." An agent that cannot book, buy or log in is a chatbot with a browser costume.
ByteHerald's reading: block, allow, or license
Site owners now face a three-way decision, and each option has a price:
| Option | What you protect | What you risk |
|---|---|---|
| Block | Ad revenue, server costs, account security | Users whose agents do their shopping simply shop elsewhere |
| Allow | Frictionless experience for agent-equipped customers | Scraping, fraud, and bots you cannot distinguish from agents |
| License (Yelp's model) | Turns the threat into a revenue line | Only works if you have data worth paying for |
The pressure runs both directions at once. We tracked the reverse side of this war last week: Meta's October 30 business terms claim the right to scrape business websites to train its AI models, while those same businesses block AI agents from transacting on their sites. Labs want the web's data in and its doors open; publishers want neither. And the agent that cannot reach a website is stuck inside your browser: ChatGPT can already read your Chrome tabs, but tabs are the easy half of the job. The hard half is the checkout button on someone else's server. This is the assume-breach world we graded in October's AI security report card: trust no agent, verify every session. The banks read that report card before the labs did.
By ByteHerald Staff. Blocking details from The Verge's October 17, 2026 report, the originator of this story; Amazon's and Perplexity's statements and Meta's response as reported there; Yelp's licensing position as told to TechCrunch and reported by The Verge. User-reported failures on Cars.com and Amtrak are unverified user reports, noted as such. Timeline graphic is ByteHerald's original compilation.
Comments
Post a Comment