Some of the largest companies in banking and retail are beginning to compete with tech giants—and each other—around a new type of customer: AI personal agents.
JPMorgan Chase, Charles Schwab, Amazon, Neiman Marcus, and Kohl's have begun blocking individual agents like Meta Muse, while consumers are just beginning to use these tools to find higher savings rates, compare prices, and cancel subscriptions.
The era of agent-based AI has arrived, but it remains uncertain how the industry landscape will shift when virtual stores are flooded with AI agents or when companies attempt to block them.
Last week’s reports about AI agents “going rogue” in customer accounts, along with prominent politicians including Senators Mark Warner and Bernie Sanders calling for regulatory legislation in response to these headlines, were followed by JPMorgan Chase CEO Jamie Dimon stating that the bank currently blocks external AI agents from acting on behalf of retail customers due to security risks.
Some retailers have also chosen to embrace this trend. Last week, Walmart announced that it would open its virtual doors to customer-preferred AI agents. In a blog post announcing the decision, the company stated: “We price products, not people, so AI agents will see the same prices as humans on Walmart.com, and applicable discounts and fees will be displayed before purchase.”
After auditing publicly accessible rules at 51 major U.S. retailers, identity verification company Billions Network found that only 11 have taken a clear stance on AI agents. Among them, five—including Amazon, Kohl’s, and Neiman Marcus—explicitly restrict AI agents, while 37 primarily block shopping agents inadvertently through outdated bot rules designed for web scrapers. None offer a mechanism for verified agents to complete purchases.
According to Billions Network's analysis, proxy blocking can be implemented in various ways: for example, like Chewy, blocking specific bots from third parties based on rules; or adopting broader restrictions to prevent any third-party bots from accessing login, registration, and checkout functions. The analysis also cautions that publicly stated rules do not necessarily guarantee technical enforcement of these restrictions.
Evin McMullen, co-founder and CEO of Billions, said that, for now, “most of the retailers in our audit have not yet made decisions regarding AI agents.”
Michael Abbott, head of global banking at consulting firm Accenture, believes that in the long term, companies that adopt a strategy of blocking AI agents will find it difficult to succeed.
Abbott said, "If a customer explicitly authorizes an AI agent to act on their behalf, it is ultimately an extension of the customer's own choices. The more important question is not whether institutions can prevent agents, but whether they can create a safe and trustworthy way for agents to participate."
Abbott stated that Accenture's research found that 65% of consumers are willing to use GPT-like financial assistants via generative AI platforms or digital wallets, and 71% are willing to use embedded AI assistants within their primary bank's mobile app.
He said, “We’ve seen this before… online banking, mobile banking, and digital wallets. Customers tend to gravitate toward experiences that reduce friction and save time. The institutions that ultimately prevail are usually those that manage to engage safely rather than resist change.”
The real opportunity for banks is to become the trusted provider of the data, intelligence, and execution capabilities that these agents rely on.
Abbott said: "As interactions increasingly occur outside of banks' own channels, banks still need to be present at those critical decision moments, even if they don't own the interface."
Give customers control over AI agents
Regarding how to distinguish between genuine security concerns and companies protecting profits by making fund transfers or price comparisons more difficult, Abbott stated that both can be true simultaneously.
He said, “There are very real concerns around fraud, authentication, accountability, and consumer protection. Customers also expect banks to safeguard their assets and data—these responsibilities do not disappear with the advent of AI agents.” He added that, at the same time, consumers have historically benefited from technologies that increase transparency and make it easier to compare options.
Abbott said, "Our research found that 82% of people want to approve every AI operation, and 79% want the ability to immediately pause or stop AI activities. This shows that consumers want convenience, but not at the expense of control."
Mike Danford, owner and chief revenue officer of Adverio, who runs Amazon, Walmart, and Target platform operations for e-commerce brands, said that some of the security concerns raised by businesses, including banks, are not entirely about security but rather about which “sandbox” customers are operating in.
Danford said, “Amazon blocks Muse and other external agents, but its own ‘Buy for Me’ feature uses Amazon’s proxies to operate on other brands’ websites. Amazon doesn’t seem to be concerned about proxies themselves, but rather about who controls the data.” He added, “Amazon now allows shoppers to set rules for automatic purchases when prices reach a threshold, and consumers appear to accept this proxy-based purchasing because shoppers still approve decisions within predefined parameters.”
An Amazon spokesperson said: "We believe that third-party apps purchasing goods on behalf of customers from other businesses should operate transparently and respect service providers' decisions about whether to participate. Third-party proxy apps like Muse have the same obligations, and we have requested that Meta remove Amazon from this experience."
Danford said that banks are similar, except they charge fees rather than build a wall.
He said, "JPMorgan Chase now charges aggregators like Plaid for accessing account data, so these costs are passed on to the applications used by consumers and may be transmitted through subscription fees."
JPMorgan declined to comment.
Plaid stated that, as of the agreement reached in September 2025, it will not affect any existing pricing arrangements with its clients. Plaid does not charge consumers directly; however, applications and services using Plaid may determine their own fees and subscription prices for customers.
Michael Jabbara, Head of Risk and Controls for the Visa Payments Ecosystem, declined to comment on any bank’s motivations for blocking agent-based AI, but stated that fraud risks from agents are “very real and very specific.”
He said that consumers may be deceived by fake agents created by scammers, leading to the theft of credentials and payment information; and even legitimate agents might perform actions that customers never intended to authorize, such as purchasing a sofa that the shopper only asked it to find.
Agents may also exhibit unpredictable behavior: Jabbara described a user who instructed the agent to book a gym time slot; when the slot was already taken, the agent hacked into the gym’s scheduling software, deleted existing reservations, and inserted the user’s name. He said that underlying concerns ultimately come down to understanding “who the agent is, who the agent is acting on behalf of,” and whether it has “appropriate permissions.”
He said, "An agent is not just about exchanging data; it can also take actions on behalf of the user."
Like other experts, Jabbara believes the shift from online shopping to mobile shopping is instructive, but for different reasons: “Before any innovation can scale, it must first establish a foundational layer of security to support trust at that scale.”
Proxy flood
Another concern is scale. Millions of agents working legally for customers could overwhelm the website and customer service systems, a phenomenon some call "agent flooding."
As personal agents like Muse suddenly surge in popularity, Wharton School AI expert Ethan Mollick posted: “We will soon see a truly legitimate distributed denial-of-service attack targeting customer service, authorized by millions of consumers; this will happen once everyone realizes that Muse/Dot and similar tools can endlessly submit old refund requests, negotiate deals, and more.”
Abbott sees this as a possibility rather than a crisis. He says, "I view it more as an infrastructure challenge than a survival threat. Most digital systems today are designed for human-to-human interactions. Agent-driven interactions may occur at entirely different scales and frequencies. One person might have multiple agents simultaneously monitoring interest rates, researching products, or evaluating alternatives."
However, he stated that the solution is not to block them, but to build the next-generation infrastructure capable of verifying trusted agents, validating permissions, managing identities, and efficiently supporting machine-to-machine interactions.
He added that this is not unfamiliar to banks.
In the past, customers interacted with banks only a few times a month, but now, with mobile platforms, they interact almost daily. AI agents are simply the next evolution. Institutions that prepare for this reality will be in a stronger position than those trying to resist it.
Although many financial websites have implemented defenses against overuse to guard against long-standing automated denial-of-service attacks, Ed Felten, Professor Emeritus of Computer Science at Princeton University and former Chief Technology Officer of the U.S. Federal Trade Commission and Deputy Chief Technology Officer under the Obama administration, says that bots "never sleep and never lose focus." Moreover, the more bots use a website, the higher the costs may become for companies such as banks.
But Felten suspects this blocking won’t last long. He says, “When a button in a user’s browser is clicked, it’s difficult for a remote website to determine whether it was clicked by a human hand or by an AI agent on the user’s computer. If there’s truly a battle over AI access, it may slow the adoption of AI agents—but ultimately, the agents will prevail because they’re so useful to users; websites that block them too effectively will eventually lose business.”
How long can the established giants keep the “door” closed?
Experts say that as market forces compel websites to accept client proxies, a shared understanding of what constitutes reasonable proxy behavior will gradually emerge.
But Joseph Chalom, CEO of SharpLink—one of the companies holding the most Ethereum and a former employee of BlackRock—expects the doors to remain closed for some time.
He said, “These large incumbent companies actually control access and will try to shut the door to fend off this threat of innovation. For decades, disruptors have been fighting for access in areas such as travel, telecommunications, and financial data aggregation.” He added that in each case, the incumbents claim it’s due to security concerns, but in reality, they also face serious threats to their business models.
“For example, AI agents working on behalf of users can easily rebook flights at a lower cost, eroding airlines’ profit margins. More importantly, AI agents can move idle cash from checking and savings accounts with nearly zero interest into money market accounts offering yields ten times higher,” Chalom said, noting that as agents gain full financial knowledge and attention, these “switching costs” are being driven down to near zero.
Therefore, I don't think they will rush to lower the drawbridge.
Abbott said that in the short term, some institutions will welcome agents, while others will maintain restrictions until issues around identity verification, accountability, and governance are resolved. In the long term, he said, data shows that people will delegate administrative tasks in financial decision-making to their agents. “Defensive strategies can keep you from losing in the short term, but to win, banks need an offensive strategy.”
Proprietary bots must be powerful enough not only to retain customers but also to make them prefer not using other agents—and this is precisely the thinking expressed by Andy Jassy and others. Jassy stated during last week’s latest quarterly earnings call: “We believe Alexa for Shopping will become the primary shopping assistant customers use when shopping on Amazon. Having a shopping assistant always available to help you browse, get inspiration, answer questions on product pages, and directly place orders based on your purchase history, shipping address, and payment information is uniquely helpful and convenient for customers.”
Ultimately, the brand is about to face an unprecedented test. Human customers develop loyalty to brands for reasons beyond price: nostalgic cereals that evoke warm childhood memories, airlines tied to family vacation experiences, or banks with familiar tellers. But AI agents don’t care about any of that.
Nikhil Behl, President of Software Business at F Token Issuance and a senior executive at the analytics company known for its credit scoring, said: “Brand loyalty and customer relationships are being tested. AI agents have no emotional attachment to a brand or a decade-long relationship—they have a task and care only about results.” He added, “The levers previously used to enhance retention will be replaced by those that agents can truly leverage, such as consistent approvals, transparent pricing, rapid issue resolution, and terms that withstand comparison.”