Multiple Platforms Restrict AI Agents from Accessing Trading and Accounts
Delta Airlines, JPMorgan Chase, Charles Schwab, and Citigroup have recently restricted external AI agents from booking tickets or operating accounts on behalf of customers. Yelp requires agents to pay for data authorization, while Amazon has banned Meta's Muse and Grok shopping agents.
Delta stated that there are currently no partnerships or integrations with third-party AI agents for purchasing or booking flights. JPMorgan CEO Jamie Dimon said that the company will not allow any external agents to access its systems due to high risks to customers and the bank, warning that there could be thousands of "rogue agents."
Charles Schwab CEO Rick Wurster mentioned that the firm intercepts dozens of unauthorized agent attempts (including Muse) to move customer funds each week, and will provide read-only balance and research access through a controlled MCP server while promoting its own assistant, Charley. Citigroup explicitly prohibits third-party agents from logging into customer accounts or trading directly.
After banning Meta Muse in September, Amazon received reports in October of Grok Bot orders being rejected, indicating that unauthorized AI agents violated the terms of service. The company previously required agents to self-identify and obtain approval, while its own shopping tools are not subject to the same restrictions.
Apollo's chief economist Torsten Slok pointed out that if agents automatically sweep demand deposits from accounts with an average interest rate of 0.1% into high-yield accounts with rates of 3.3%-5%, large banks could face substantial interest costs. JPMorgan, Bank of America, and Wells Fargo hold about $16 trillion in non-interest-bearing deposits, which could increase annual interest expenses by approximately $47 billion at a 3% interest rate.
E-commerce platforms are concerned that agents may bypass advertising and recommended traffic, affecting advertising revenue and customer relationship control. Currently, institutions generally refuse to hand over account control to external agents until security issues like prompt injection are resolved.
Market Mechanism: Platforms and banks use technical blocks and policy restrictions to protect their customer relationships and interest income, redirecting funds and traffic to their own agents or paid authorization channels, benefiting the platforms that control accounts and inventory, while putting pressure on users and agent developers who rely on third-party agents.
Source: Public Information
ABAB AI Insight
Amazon has sued Perplexity and banned its shopping agents between 2025-2026, subsequently taking similar measures against Meta Muse, requiring agents to self-identify and obtain approval. On the banking side, there has been a long-standing cautious attitude towards third-party credential logins under the open banking framework, with executives like Dimon explicitly listing "rogue agents" as a risk during earnings calls.
In terms of capital pathways, banks protect low-cost demand deposits by intercepting external agent logins to avoid increased interest expenses from automatic sweeps (Apollo estimates that only a few large banks could face hundreds of billions in potential costs), while promoting their own assistants and read-only interfaces. E-commerce platforms use terms and technical means to keep transaction traffic on their own interfaces, preventing advertising and recommendation revenue from being bypassed.
This is analogous to the disputes over bank fees caused by early open banking and data intermediaries like Plaid, or the restrictions airlines place on third-party booking engines. The industry is currently in a phase of rapid expansion of agent capabilities, but the security and responsibility framework has yet to be established, leading institutions to choose to block first and build later.
Structurally, this represents a transfer of pricing power and capital concentration: if external agents can automatically optimize deposits and bypass interfaces, it will weaken banks' interest margins and e-commerce advertising pricing power. The mechanism is to reduce user switching friction and bypass platform recommendation layers, eroding the profit structure of "lazy funds" and "interface control," thus forcing platforms to temporarily maintain control through blocking.
ABAB News · Cognitive Law
Agents can optimize, platforms block first
Low-cost deposits rely on inertia, automatic optimization is a threat
Bypassing interfaces equals bypassing profits