Meta (META.US) Muse explosion raises financial concerns Bernstein: Banks, insurance, and brokerage business models face reshaping Visa (V.US) and Mastercard (MA.US) or become beneficiaries

The Zhitong Finance App learned that Bernstein said in the latest report that Muse, the consumer-grade AI smart device launched by Meta (META.US), may become an important turning point in the consumer AI field. Since its launch, Muse has quickly reached the top of the US app store. Currently, it has reached 2.8 million downloads, and has begun to be used by consumers for a series of tasks such as making reservations, ordering meals, tracking ticket prices, canceling subscriptions, comparing insurance products, filling in forms, and contacting customer service. Bernstein believes that it is only a matter of time before other cutting-edge AI models and technology companies follow up and launch consumer-grade AI agents; a “consumer AI smart device competition” is taking shape.

This change has quickly been reflected in financial markets. Bernstein pointed out that the rapid popularity of Muse has raised investors' concerns about business models that rely on consumer usage habits and shifting frictions. Since the launch of Muse, stock prices in some financial sectors, such as insurance, brokerage firms, banks, and mortgage institutions, have dropped by a cumulative total of 5%-15%.

However, Bernstein believes that some of the “apocalyptic scenarios” that currently revolve around AI agents disrupting the financial industry in the market are too simple. What really determines whether AI agents can transform the financial services industry on a large scale is not whether technology can complete these tasks, but whether consumers are willing to hand over financial decisions to AI, whether financial institutions allow them to access data, and whether accountability and regulatory rules can keep up.

AI agents directly hit the traditional “moat” of the financial industry, and banks, insurance, and brokerage firms bear the brunt

The biggest difference between AI agents and traditional chatbots is that they can not only provide suggestions, but also act on behalf of consumers.

Bernstein pointed out that an AI intelligence that can automatically shop, compare prices, change service providers, negotiate, and even transfer funds may gradually weaken a series of user habits and conversion barriers that have benefited the financial industry for a long time, including long-term use of the same credit card, stickiness in bank deposits, idle cash balances at brokerage firms, and insurance renewal pricing.

Take insurance as an example. In the past, consumers often didn't actively compare quotes from different insurance companies every year. If AI agents can automatically complete price comparisons and help consumers change insurance companies, some of the pricing advantages that insurance companies gain by relying on customer renewal inertia may be weakened. Similar logic applies to banks and brokerage firms. AI agents can continuously search for cash management products with higher returns, making it easier to transfer funds between different accounts, thereby reducing the revenue that brokerage firms receive from customers' idle cash. For banks, easier deposit migration may drive up capital costs and put pressure on net interest spreads.

At the same time, AI agents that can automatically monitor portfolios and perform financial planning may also reduce consumer demand for some paid manual financial management services.

The credit card industry is also likely to be affected. If AI can decide which card to use in real time based on rewards, interest rates, and offers, consumers' long-term habit of using a credit card as a “preferred card” may gradually weaken.

The willingness of institutions to open up is a key constraint

Despite the huge potential impact, Bernstein believes that third-party AI agents still face a core contradiction, that is, AI agents are difficult to spread on a large scale without the participation of merchants and financial service providers; however, if participation means losing checkout entrances, customer relationships, or some financial benefits, these companies have no incentive to fully open up.

Banks, brokerage firms, insurance companies, and payment companies still firmly grasp key aspects such as account login, identity verification, official quotation, and product qualification review.

Amazon (AMZN.US) blocking Muse is a typical example. According to the report, Amazon believes that Muse did not clearly identify itself as an AI agent and obtained customer login credentials, but Muse has different opinions about this. The online insurance platform Insurify also blocked Muse on the grounds that AI automatically fetches insurance quotes and may only show prices, but omit information disclosed in insurance amounts, deductibles, eligibility conditions, and regulatory requirements.

Bernstein predicts that before the consumer AI ecosystem finally matures, there will be more cases of blocking AI agents, paid data access agreements, and financial institutions developing their own AI agents. The complete opposite business model may even emerge: instead of AI platforms charging financial institutions, banks charge AI agents for data access.

According to the report, financial institutions still have access to key entrances such as customer authentication, account access, and data sharing rights. Previously, J.P. Morgan Chase had already begun charging data aggregators for customer data access fees, which provided a reference for financial institutions to charge AI platforms in the future.

Consumers prefer to let AI “assist” rather than “make decisions” trust is still a core challenge

Bernstein believes that the financial services industry is based on “trust,” and this is one of the most obvious shortcomings of current consumer AI devices.

According to TD Bank's 2026 survey of more than 2,500 American consumers, 55% of respondents are already using AI to help manage their finances, compared to just 10% a year ago. However, while 62% of consumers believe AI can provide reliable information, only 18% are willing to let AI make important financial decisions independently.

Consumers are more inclined to let AI improve efficiency and convenience, while humans still have the final say. Other surveys have shown similar results. According to a survey of American and British consumers conducted by ACI Worldwide and YouGov in June of this year, only 7% of consumers are willing to let AI assistants shop directly without their approval; an Accenture survey of 25,590 consumers showed that 32% are willing to let AI make purchasing decisions within pre-set limits, but only 12% are willing to let AI make completely autonomous decisions in the payment process.

Bernstein believes that AI agents are more likely to act as “assistants” in the short term rather than completely take over consumers' financial lives.

AI agents go deep into financial scenarios, and the regulatory framework still needs to be improved

Another major difference between financial services and ordinary e-commerce is that it is a highly regulated industry. Bernstein pointed out that AI technology is currently developing at a much faster pace than the accountability and regulatory framework. As AI agents begin to apply for loans, buy insurance, make payments, and even manage investments for consumers, a range of unresolved issues will become increasingly important.

For example, if AI submits loan applications or purchases financial products on behalf of consumers, does it constitute a valid authorization? How can banks confirm that AI actually represents real customer actions? Under what circumstances would AI provide product optimization recommendations be considered regulated financial advice? If AI compares or even directly purchases insurance, does it need to obtain a relevant license? If AI uses outdated data, acts beyond the scope of consumer authorization, or selects products that are highly inappropriate for customers, who is ultimately responsible?

Bernstein predicts that traditional financial institutions will not passively wait for AI agents to steal customers. In addition to limiting data access and promoting stricter regulation, many companies may launch their own AI agents, or actively cooperate with third-party AI platforms, or even sacrifice some of their existing revenue to reduce the risk of customer churn.

Instead, the payments industry may be winners Visa and Mastercard are favored by Bernstein

Notably, unlike market concerns that AI agents will disrupt banks, insurance, and brokerage firms, Bernstein's judgment on the payment industry is clearly more optimistic.

According to the report, AI smart businesses are generally viewed as a potential threat to Visa (V.US) and Mastercard (MA.US), but the actual situation may be “just the opposite.” As AI agents bring more digital transactions, and demand for value-added services such as identity authentication and risk management grows, payment networks may instead become beneficiaries.

Bernstein believes that the early stages of AI smart business development may be similar to the early stages of e-commerce, and will also face problems of fraud, consumer confusion, and lack of trust. Visa and Mastercard essentially provide a “trust network” whose advantages are not only payments themselves, but also dispute handling, transaction standards, risk management, and a huge global acceptance network.

According to the report, the two companies have more than 7 billion payment vouchers and cover more than 100 million merchants. As a result, Bernstein believes that in the age of AI smart devices, credit cards may still become the main payment method.

As the complexity of AI agent transactions increases, the “tokenization” of payments may also become more popular. Products such as Meta Muse have already begun to use disposable virtual cards, while Visa and Mastercard have also launched payment solutions for AI agents, such as Visa Intelligent Commerce and Mastercard Agent Pay, respectively.

As of September 25, Bernstein gave Visa and Mastercard “outperform the market”, with target prices of $450 and $710, respectively; Adyen and Affirm were also given “outperforming the market” ratings, with target prices of 1,600 euros and 110 dollars, respectively.

PayPal is in a more delicate situation, and BNPL may have an advantage

The future of digital wallets is more complicated. Bernstein pointed out that if consumers let AI agents directly complete payments in the future, then friction such as “tourist checkout” in traditional e-commerce may become less important, which may weaken part of the traditional value of digital wallets.

But on the other hand, bilateral digital wallets like PayPal (PYPL.US) also have the opportunity to expand their value in terms of trust, security, fraud and transaction dispute handling, while helping consumers automatically find offers, rewards, and optimal payment methods.

The report also suggests that as Google continues to improve the Spark feature in Gemini, it is worth paying attention to whether deeper cooperation between Google and PayPal may occur in the future.

Meanwhile, Bernstein believes that “buy now, pay later” (BNPL) platforms such as Affirm (AFRM.US) and Klarna may be in a relatively favorable position, particularly in terms of interest-bearing loans, zero-interest loans, and transparency of loan terms involving subprime consumers.

AI won't disrupt the financial industry overnight; trust, data, and regulation are the keys

Overall, Bernstein doesn't think the explosion of Muse means that the traditional financial industry will quickly be replaced by AI agents. The report points out that in the early stages of development, AI agents are more likely to take the lead in resolving 90% of friction in complex processes, such as booking services, comparing insurance products, or handling travel disruptions, rather than immediately redefining daily retail checkout or traditional financial services.

In the financial industry in particular, Bernstein believes that the three factors that really limit the large-scale adoption of AI agents are not model ability, but consumer trust, data access, and regulatory rules.