Delta, JPMorgan, and Schwab blocking AI agents is no glitch

Delta won't let an AI agent book your flight. JPMorgan Chase and Charles Schwab have confirmed they're blocking agents from touching accounts entirely. Yelp will let them in — for a licensing fee. If you've been planning to automate workflows through AI agents acting across the open web, this week's news should make you stop and recalculate. The agentic internet isn't going to be open. It's going to be gated, licensed, and commercially negotiated, and most mid-market businesses won't be at the table when those deals get made.

According to The Verge, users of Meta's Muse, xAI's Grok, OpenAI's Dots, and Instinct have all hit walls in recent weeks, with agents failing to complete tasks across major websites. JPMorgan and Schwab cited security concerns. Yelp was more direct: it told TechCrunch it only allows AI agents that pay for its data licensing program. Delta is blocking ticket purchases outright. Users are also reporting that Cars.com and Amtrak — sites that apparently worked before — have gone dark for their agents, though neither company has confirmed this is deliberate policy.

The legal backdrop makes the commercial logic even clearer. Amazon sued Perplexity last year over its agentic shopping feature, claiming it had 'repeatedly requested' that Perplexity stop sending agents to its site. A judge initially ordered Perplexity to comply; an appeals court overturned that ruling in August, but Amazon has since filed an amended complaint. Amazon also told Meta that Muse's 'continued access' violates its site policies. Perplexity's read on Amazon's motive is blunt: the company doesn't want agents on its platform because agents don't look at ads.

This isn't a friction problem — it's a structural one

Here's the thing about framing these blocks as 'security concerns' or 'technical conflicts': some of them genuinely are. Chris Canal, CEO of AI safety evaluator EquiStamp, told The Verge that '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.' That's real. Anti-fraud systems can't tell a legitimate agent from a credential-stuffing script, and a bank blocking unfamiliar automated access to customer accounts is doing exactly what we'd want it to do.

But security concerns don't explain Yelp's licensing program. They don't explain Amazon's lawsuit. They don't explain Delta deciding that AI agents simply can't book tickets. Those are business decisions. The companies blocking agents aren't confused — they're negotiating. They want to control who gets access, under what terms, and for how much money.

Meta seems to understand this perfectly. Rather than fight the blocks, it's cutting deals: Walmart, Gap, Sephora, and Best Buy have all partnered to let Muse shop on their sites. Shopify has opened its storefront network to agents. This is what the agentic web actually looks like — not a free-range system where your agent wanders anywhere you point it, but a patchwork of commercial agreements between AI platforms and the sites willing to participate. Even that patchwork is leaky: Reddit reports suggest Google is flagging some accounts using Muse for 'suspicious activity,' and users are hitting checkout failures on Walmart despite its Meta partnership.

Elon Musk's response to Amazon blocking agents was to float starting a competitor. That tells you everything about where this ends up: a fragmented ecosystem where each major AI platform negotiates its own access deals, and anyone not running one of those platforms is dependent on whatever coverage their vendor has managed to arrange.

The fair counterargument is that this is early-market friction, not permanent architecture. Standards bodies could establish agent authentication protocols. Regulators could step in. The web didn't stay closed to mobile apps forever. That's not wrong — but it misses the timeline. Businesses automating workflows on agents right now are betting on access that can be revoked by a policy update, a lawsuit, or a licensing negotiation they have no part in. That's not a calculated risk. It's an unexamined one.

Canal put the stakes plainly: 'If there are companies that are actively working to make that not happen, there are just fewer useful things the agents can do.' That's a direct hit on the core pitch from Anthropic, OpenAI, and every other vendor selling agentic AI as the thing that will 'permeate every aspect of our lives.' Fewer sites, fewer tasks, less value — and the agent vendors are the ones absorbing the reputational cost while the blocking companies absorb none of it.

What this means if you're running a business

First, if a vendor has sold you on AI agents handling external workflows — booking travel, pulling financial data, interacting with partner platforms — ask them this week which specific sites their agents can reliably access, and what happens when access gets cut off. 'We're working on it' is not a service-level commitment.

Second, any internal automation that touches third-party sites should be audited for fragility. An agent that books Delta flights or pulls data from a financial platform is one policy update away from breaking. That's a workflow dependency you should know about before it fails on a deadline.

Third, watch your vendor's partnership announcements carefully. Meta's deal with Walmart and Shopify's open network are the model for how access will be structured. If your AI platform of choice isn't cutting these deals, your agents will have shrinking reach over time — not growing reach.

The picture Canal paints of CAPTCHA prompts appearing on grocery sites to screen out agents sounds absurd. It probably isn't. The web has been through this before with bots, scrapers, and automated checkout tools. The pattern is always the same: access gets restricted, then monetized, then tiered. Agents are just the latest thing going through that cycle.

Watch for whether OpenAI and Google — both of which hadn't responded to The Verge's questions at publication — announce their own site partnership programs in the next few months. If they do, that confirms the licensed-access model is winning. If they push back legally or technically, there's still a fight to be had. My bet is on the licensing deals. The companies doing the blocking have too much revenue at stake to leave access unmonetized.