Heard on the Street’s Asa Fitch opens with a sentence Meta’s communications team will not enjoy: “User trust isn’t Meta Platforms’ forte. Yet the company’s AI future increasingly depends on it.”
Last week Meta (META) rolled out Muse, which Mark Zuckerberg calls a “personal AI agent built for everyone.” The agents inside Muse can book a doctor’s appointment, buy products, manage finances and send email on your behalf. To do that, they need your email, your calendar, your shopping accounts and your bank login. An Oppenheimer survey of American consumers found that 8% would trust Meta with their passwords. Less than a third as many as would trust Google (GOOGL). Goldman’s Eric Sheridan called data-privacy hesitancy “likely the biggest potential headwind for broad-based adoption of consumer AI agents.”
The record people remember
Fitch runs the ledger without editorializing. A record $5 billion FTC fine in 2019 over Cambridge Analytica. 533 million users’ information leaked the following year. A $1.3 billion European fine in 2023 over data transfers. An $18 billion settlement over harm to teenagers. Against that, Meta announced Muse as “safe, secure, private,” running on a walled-off virtual machine no other agent can reach, and says it delayed the launch for months to work on safety. Zuckerberg dismissed the idea of slowing frontier development alongside Anthropic, OpenAI and SpaceX, arguing the incentives to keep agents in line are natural.
And then the registration screen: the agent “may make mistakes or take unexpected actions, so review its work.” An agent that buys a $1,000 item when you wanted the $10 one is a very specific kind of mistake, and it is the kind that ends the relationship.
The arithmetic
Muse is free for light use and $20 or $100 a month for heavier users. Oppenheimer estimates that to lift earnings per share by roughly 20%, Meta would need 115 million subscribers paying $20 — about double the number of people who pay for ChatGPT today. That is ambitious for anyone; it is more ambitious for the company with the trust deficit. Meanwhile Meta is spending an expected $137 billion on capital expenditure this year, funded by an advertising business that has used AI brilliantly, and Muse is one of the cogs meant to earn a return on it. Investors have given it the benefit of the doubt: the stock is up about 8.5% since launch and closed Friday at $665.75.
Our read
This is the third agent story in three editions — the Mastercard card issued to your robot on Friday, the check-back question in every one of our letters this month, and now the agent that wants the bank login. The desk’s position has not moved: own the platforms, because the transaction layer is where the money lands, and Meta is held in the tactical books for that reason. But the survey is a real number, and it argues that the winner of the agent race is the company whose email and calendar you already use, not the company asking you to move them. Alphabet is held at a larger weight than Meta, and this column is a decent explanation of why.
The household rule is unchanged from Friday and worth repeating because the product just got more capable: nothing that moves money, nothing that signs, and leave the check-back turned on. Meta wrote the warning itself.
