Three kinds of no
OpenAI, Meta, Google and Instinct all promote agents that scour the internet on a user's behalf — booking reservations, clearing an inbox, buying shoes. In recent weeks, users of Meta's Muse, xAI's Grok, OpenAI's Dots and Instinct's agent have reported blocks stopping those agents from completing tasks on major websites. The Verge's October 17 survey collected the websites' answers, and the answers are not one policy repeated. They are three different decisions, taken for three different reasons.
Yelp's answer is commercial. The company told TechCrunch that it only allows AI agents that pay for its data licensing program. Read precisely, that is a statement about how access is granted: through licensing arrangements negotiated with Yelp. It is not a published price list that any agent can pay for immediate, unrestricted entry, and it does not make Yelp's data open to whichever agent shows up with a card.
Delta Air Lines' answer is narrower, and its boundary is unknown. The reporting establishes that the airline is preventing agents from booking tickets on behalf of users. It does not establish what Delta thinks of anything short of a booking — reading schedules, comparing fares, holding an itinerary for a human to finish. None of that is confirmed either way, and the booking restriction should not be inflated into a general ban on agents touching Delta's site.
The banks' answer is about account security, and the two banks should not be merged. JPMorgan Chase confirmed that it is blocking AI agents from accessing customer accounts, citing security concerns. Charles Schwab confirmed the same kind of block. The fuller explanation in the reporting comes from Schwab alone: its chief executive, Rick Wurster, told Bloomberg — in an interview reported by The Verge on October 16 — that he does not want a client handing a password to Muse and letting it act on the account, because the bank cannot control it, the client cannot control it, and if something goes wrong it is unclear who is liable. That reasoning is Wurster's, speaking for Schwab. JPMorgan gave no equivalent explanation in this reporting, and its block should not be dressed in Schwab's words.
The lawsuit underneath
Amazon has been running the legal version of this argument for a year. It banned Perplexity's agentic shopping feature, said it had repeatedly asked the startup to stop, and sued. Amazon's position is that third-party companies buying on behalf of customers should operate openly and respect a service provider's decision on whether to participate. Perplexity's answer is that Amazon's real objection is advertising — an agent does not look at ads — and that its agent is an extension of the user, like an assistant or employee acting for them.
The court sequence matters, and it has not produced a winner. A judge ordered Perplexity to stop its agents shopping on Amazon. An appeals court overturned that order in August 2026. Amazon has since filed an amended complaint. The dispute remains unresolved: the reversal removed the order, it did not settle the underlying claims, and neither company has a final victory.
The fight has also spread beyond Perplexity. Amazon has told Meta that Muse's "continued access" to Amazon violates its site policies, per The Verge, and there are reports of Grok, Instinct's agent and OpenAI's Dots being unable to use Amazon. Elon Musk's response, per The Verge, was to float starting a competitor to Amazon.
Deals are one way through
Meta's public response to being blocked is that "turning away a personal agent means turning away the customer behind it". Rather than fight Amazon, it has partnered with Walmart, Gap, Sephora, Best Buy and others so that Muse can shop on their sites with permission. Shopify has also opened its network of storefronts to AI agents.
That makes partnerships one working route around the blocks. It does not make access a settled right. A storefront network opening its doors is not proof that every merchant on it accepts every agent without conditions, and the arrangements differ by provider: two agents built on similar models can end up with very different usefulness because one maker signed the retailer and the other did not.
What has not been established
Two of the names in the survey sit in a weaker category, and they should stay there. Cars.com and Amtrak are user reports only: users say access that previously worked with their agents no longer does. Neither company has confirmed a deliberate policy. Treating those reports as confirmed bans would overstate the reporting.
Some failures may not be decisions at all. Agents trip anti-bot systems built to catch scammers and scrapers. Chris Canal, who runs the AI safety evaluator EquiStamp, told The Verge: "There are all kinds of things that are set up not even to block agents, but to block in general bots or programs that are trying to scam or hack. These AI agents trigger those same safeguards." Reddit reports in the piece describe Google flagging some Muse accounts for "suspicious activity", and users having trouble checking out at Walmart despite the Meta partnership. Canal also made the value argument directly: if companies are actively working to stop agents doing what users can do online, there are simply fewer useful things the agents can do — and the defensive response, more human-verification screens, will make ordinary websites more annoying for everyone.
OpenAI and Google did not immediately respond to The Verge's requests for comment, and Instinct declined to comment. Their silence is not agreement with any side's account of these blocks.
The practical test
If you are deciding whether an agent subscription is worth paying for, start with the paperwork, not the demonstration. Check the agent's documented integrations and permissions — the services its maker says it is authorized to work with — and then test the tasks you would actually delegate: the booking, the order, the account you care about.
Treat the results of that test as provisional in both directions. A task that succeeds today is not guaranteed to keep succeeding; permissions, partnerships and fraud controls change. A task that fails is not proof of a formal ban; it may have hit an anti-bot safeguard that was never aimed at agents. And for bank accounts specifically, prefer an authorized integration the bank itself offers over handing a general-purpose agent your account password — the password handover is precisely the arrangement Schwab's chief executive described rejecting.
Sources
- [1] The Verge, Emma Roth, "Websites are closing their doors to AI agents," October 17, 2026 — the survey, the company confirmations, the Amazon litigation sequence, the retail partnerships and the EquiStamp commentsRead source
- [2] The Verge, October 16, 2026 — item on Charles Schwab blocking AI agents, quoting CEO Rick Wurster's Bloomberg interviewRead source