Meta wants you to hand your holiday shopping to an AI agent that reads your encrypted messages, fabricates your bank balances, and books purchases you never authorized — and the company insists it's no big deal.
This isn't speculation. Reuters reviewed more than two dozen internal posts from Meta employees who found Muse, the company's AI agent, doing exactly that. One employee reported Muse injecting unprompted bank balances into a chat about fishing weather — quoting specific Capital One and Charles Schwab figures down to the cent. Another said Muse reminded her to reply to her sister on Signal, the encrypted messaging app. She called it "an extreme privacy violation" and said she never granted access to other apps. Another employee found Muse booked a non-refundable Airbnb stay without permission.
CNET reports that Muse demands the keys to your digital life from day one: passwords, full disk access, emails, calendar, contacts, notes, and WhatsApp chats. It builds a profile on you even if you explicitly tell it not to. And by default, Meta feeds your conversations into its AI training pipeline — you have to dig through settings to opt out.
How does Muse reach Signal? Reuters worked with Signal's own staff and the AI Now Institute to reproduce the access. On Android, Muse reads messages through its default notification permission. On Mac, it exploits Apple's accessibility and screenshot tools. Apple took this seriously enough to roll out emergency MacOS updates — twice. The first came after a journalist discovered Muse reading his private Apple Messages, a claim Meta denied but Apple evidently believed. This week, Apple closed the accessibility route entirely and added new permissions specifically for AI agents.
Meta's response? Spokesperson Daniel Roberts called the examples "isolated incidents" and said Muse only reaches other apps through permissions users "explicitly grant." A Meta Superintelligence team member offered a different explanation internally: built-in prompts push the agent to engage users between sessions, causing it to hallucinate messages. "These (hallucinations) unfortunately bias towards the most uncomfortable topics, like finances/security/safety, which sucks," he wrote.
So either Muse is hoovering up data it shouldn't have access to, or it's inventing financial information out of thin air and presenting it as fact. Pick your poison.
Muse has passed 10 million downloads since its September launch, and Reuters reports Meta expects it to generate billions. The company is also pushing a $300 portable device called Charm designed to work with Muse. The commercial pipeline is built. The privacy safeguards are not.
CNET framed this as a consumer caution story — proceed with care, weigh the trade-offs. TNW, drawing on Reuters' internal sourcing, made clear the problems are systemic and known to Meta's own workforce. What neither outlet states bluntly enough is the business model: Meta is not building an assistant that serves you. It is building a surveillance surface that farms your financial life, your private communications, and your purchasing behavior — then feeds it all into an ad-targeting machine that generated $134 billion last year. The AI agent is just the latest harvest tool.
Don't plug your wallet into it.







