Amazon, Delta, and the Banks Are Locking Out AI Agents — And the List Is Growing

by Geth Author

Over October 17–18, 2026, a wave of major companies confirmed they are blocking third-party AI agents from their services. Amazon, Yelp, Delta Air Lines, JPMorgan Chase, and Charles Schwab — each independently, and without a coordinated industry standard to follow — have all begun restricting autonomous AI tools from operating on their platforms.

The full picture was assembled by The Verge, which reported confirmed blocks from all five organizations in a span of 24 hours. The affected agents include Meta's Muse, xAI's Grok, and OpenAI's Dots and Instinct.

What each company is doing

Amazon confirmed it has blocked Meta's Muse agent since September 20. A spokesperson told reporters that Meta "never disclosed that Muse would access its site," and Amazon has formally asked Meta to remove Amazon from Muse's agent experience. An ongoing lawsuit between Amazon and Perplexity AI over unauthorized agentic shopping adds legal weight to the confrontation.

Yelp has limited automated access to paying customers only. Any AI agent outside its paid data licensing program is now cut off from the company's restaurant and business directory — effectively making Yelp invisible to AI agents that don't pay for the privilege.

Delta Air Lines is preventing outside AI agents from booking flights on its behalf, citing security concerns and the absence of any formal arrangement with AI providers. The block applies to all third-party agents regardless of the AI company involved.

JPMorgan Chase and Charles Schwab are keeping agents out of customer financial accounts. Both banks cite security. Their blocks extend to any AI assistant attempting to automate account actions — checking balances, transferring funds, or placing trades — on behalf of a user.

Why this is a structural problem

The agentic AI model depends on assistants being able to reach services on users' behalf: booking a flight, checking a restaurant, placing a trade. If those services actively block agents, the AI's core value proposition collapses. The user still pays for the subscription; the AI still tries to help; the destination simply refuses to answer the door.

There is no legal or technical standard governing this. Companies are responding one by one, adding IP blocks, bot-detection heuristics, and terms-of-service clauses as they go. The result is a fragmented landscape where AI capability depends not on what the model can do, but on which services decide to permit access on any given day.

The authorization gap

At the heart of these conflicts is an authentication and authorization gap. When a user asks an AI agent to book a flight, the airline has no way to verify that the user actually consented — or that the agent is acting in good faith and not scraping data, harvesting prices, or bypassing checkout flows that carry revenue.

A voluntary industry standard — something analogous to OAuth for agent authentication — is being discussed but does not yet exist. Until it does, every new AI agent capability that touches a third-party service is effectively operating in a legal and technical grey zone that the service can close unilaterally at any time.

What's next

The next 12 months will likely determine whether the agentic AI industry wins cooperation through standardized APIs and revenue-sharing data licensing, or whether the default becomes an adversarial dynamic where agents and services race to outmaneuver each other.

For now, the message from Amazon, Yelp, Delta, and the banks is clear: if an AI agent wants access, it needs to ask permission first — and permission is not currently being granted.