Meta's Muse Finds Real Cash, Hits Ad-Model Trust Wall

The reporter who tried Meta's new personal agent, Muse, actually got a check out of it. On TechCrunch's Equity podcast, Sean O'Kane said Muse scanned for unclaimed funds on his behalf and turned some up. Then he deflated the demo, calling it "a party-trick type thing" rather than the kind of task that would keep him coming back.

The bigger promised features (cancel subscriptions, flag duplicate charges, sit on top of a credit card and Gmail) need the account access he does not want to hand over. His reason was flat: "Meta's business is to sell you ads." He said he would trust Apple more with that kind of sensitive financial information because "its business is not to sell me a bunch of crappy ads."

Meta's answer is technical. Each user's agent, per the company's account, runs in a dedicated cloud computer, cannot see real credentials for the services it connects to, and is gated by a separate system that decides whether to allow, deny, or prompt an action. Conversations and tool-use records may train future Meta models unless the user opts out. The Confidential VM meant to cryptographically block Meta itself from that cloud data is not yet available.

Co-host Kirsten Korosec framed the pitch generously, pointing to Meta's record of "embedding themselves in everyday people's lives" through Facebook, Instagram, and WhatsApp. Anthony Ha wondered whether the consumer lane, away from where OpenAI and Anthropic are pointed, might be Meta's real opening. It slots into a busy stretch on our Meta tracker; Wired ran its own look at the adults-only assistant hiding behind a plush mascot called Jolly the same day.