Muse, Meta's new AI agent, sparks Wall Street fears of a bank deposit flight

By InvestorPlace · September 29, 2026.

Since its launch on September 8, Muse, Meta's (META) artificial intelligence (AI) agent, has held the number one spot on both the App Store and Google Play. According to Sensor Tower data cited by InvestorPlace, downloads have already passed 3.4 million, a faster pace of adoption than ChatGPT's in its first weeks. JPMorgan reckons Muse has the potential to become the most widely used consumer AI app since ChatGPT.

What matters is not that Muse converses better than a chatbot, but that it acts. It is an agent in the technical sense of the term: a system that does not merely answer questions but carries out tasks on its own, continuously and coordinating several subprocesses ("sub-agents") at once, with access to Gmail, PayPal, Shopify, Walmart, Best Buy, Expedia and, through the financial connector Plaid, to the user's bank accounts. Amazon has already shut the door on it, blocking its access to the platform.

That ability to act autonomously and without emotional attachment is, according to the argument the investment firm lays out, a commoditization machine: an agent does not recognize a logo, does not fall for a "limited time" offer and does not forget to cancel a free trial; it simply compares price and value. A Checkout.com survey cited in the article puts at 57% the share of consumers willing to let a shopping agent switch brands if it finds better value for money; another, from Deloitte, puts at 44% the share of retail executives who expect AI to weaken brand loyalty. Bain & Company warns that shopping agents could reduce retailers to mere order fulfillers ("drop shippers").

The sector most exposed in the short term, however, is banking. Torsten Slok, chief economist at Apollo Global Management, raises in a report cited by InvestorPlace the possibility of an "agentic bank run": if households use AI agents to optimize the return on their cash, banks could lose much of the cheap deposits they use to fund their loans. The reason is simple: traditional banks pay an average of 0.1% on checking account cash in the US, while fintechs such as SoFi Technologies (SOFI) offer 4.5% (with conditions), and Slok calculates that agents like Muse could move that money into accounts paying between 3.3% and 5.0% with barely a few seconds of instruction to the agent.

The market has already priced in that fear. The SPDR S&P Regional Banking ETF (KRE) is down close to 10% from its recent highs, and in a single recent session the S&P 500 financial index fell more than 2%, with Charles Schwab (SCHW) and other brokerage firms retreating more than 5%; InvestorPlace attributes that particular move to fear of Muse, not to interest rates, although the simultaneous rise in the 10-year bond yield had already been putting pressure on the sector in its own right.

The scale of the alarm deserves some qualification: InvestorPlace's own article describes the threat as a genuine but still theoretical risk. Agents have yet to prove that they buy or manage money better than a person at scale, and banks have room to react by launching their own AI assistants to retain cash. Amazon's resistance to Muse points in the same direction: the companies with the most to lose are not going to sit still.

On top of the market disruption comes a political front. On Sunday, Senator Elizabeth Warren (Democrat of Massachusetts), together with Tina Smith, Jeff Merkley and other colleagues, sent letters to the chief executives of Meta, Amazon, Alphabet (GOOGL) and Microsoft (MSFT) demanding explanations for the tax breaks they have applied to their AI and data center spending under the 2025 tax law known as the "One Big Beautiful Bill". According to the figures set out in the letter itself, Meta's federal tax on profits fell from $9.6 billion in 2024 to $2.8 billion in 2025 with practically identical earnings; Microsoft's tax bill dropped by more than $11 billion between fiscal years 2025 and 2026; Amazon's by almost $8 billion; and Alphabet's by more than $7 billion. The US Congress's Joint Committee on Taxation estimates at some $67 billion the retroactive deductions available in 2026 alone. Warren accuses these companies of having cultivated their relationship with the Trump administration to secure that treatment; this is a political allegation we attribute to the senator, not a proven fact.

Our reading is that the Muse episode is the most concrete evidence yet of a thesis we have been arguing for months: agentic AI does not destroy value, it redistributes it, and it does so first wherever margins depended on customer friction rather than on any real advantage. Cheap deposit banking, insurers that overcharge anyone who does not shop around at every renewal, credit cards with fine print, telecoms with a "loyalty penalty": that entire business model built on human laziness now has, for the first time, an adversary that does not tire, does not get emotional and does not forget to compare. In the short term that means real pressure on margins and probably on jobs in branch networks and administrative areas of banking and insurance, something we have already been flagging in our sector analysis. In the long run, however, it is exactly the kind of force that underpins our core thesis: an agent that works tirelessly in the ordinary user's economic interest, and not just the institution's, is an early and very concrete form of the abundance AI promises to distribute. The question that really matters is not whether Muse wins or loses this particular contest, but who controls the toll on that constant inference and who ends up on the wrong side of the friction that disappears.