Meta's consumer AI agent, Muse, has surpassed 10 million downloads since its September launch, but a Reuters investigation has uncovered privacy concerns that could shape how the tech giant—and the broader industry—handles user data. The report found cases where broad device permissions allowed the bot to pull in message text and sometimes mix up context between accounts, raising questions about what AI agents can see and how they use it.
What the investigation found
Reuters reviewed internal staff posts describing "undesired actions" from Muse, including made-up purchase details and odd, out-of-context messages that employees feared could point to cross-account data bleeding. One example appended supposed Capital One and Charles Schwab balances; another showed a reminder tied to a Signal message that an employee called an "extreme privacy violation."
Working with Signal staff and the AI Now Institute, Reuters reproduced a way for Muse to read Signal text without Meta's official "Connector" feature, by relying on setup permissions instead: push-notification access on Android, and on macOS a mix of accessibility and screenshot tools. The same request was blocked on an iPhone, and Apple restricted the pathway Reuters identified while adding more agent-specific permission prompts.
Meta spokesperson Daniel Roberts told Reuters the issues were isolated and that many mix-ups were hallucinations that have since been addressed. The bigger point: consumer "agents" are quickly forcing operating systems to redraw what apps can see.
Why this matters for your money
For everyday investors, the story is less about Muse itself and more about what it signals for the AI industry. Meta is betting heavily on AI to drive future revenue, and Muse is central to that push. The app has stayed high in download charts after promotion across Facebook, Instagram, and WhatsApp, and Meta has said about a quarter of new users return daily. That kind of engagement is valuable, but privacy missteps could invite regulatory scrutiny and erode user trust—both of which can hit a company's bottom line.
For users, the practical takeaway is about permissions. End-to-end encryption helps keep messages private as they move between apps, but it can't hide what your device displays. If an AI agent gets broad permissions at setup, it may capture plaintext from endpoints like Android notification previews or whatever macOS accessibility and screenshot tools can "see," even without an in-app connection.
That's why iPhone-level blocking, and Apple's new granular prompts for agent access, matters. If Apple and Google tighten defaults further, you may face more permission pop-ups for automations, but fewer paths for a helper bot to ingest or repeat sensitive snippets.
What investors should watch
For investors, the key is to monitor how Meta and other tech giants handle these issues. Regulatory responses to AI privacy concerns could affect not just Meta but the entire sector. Companies that fail to protect user data may face fines, lawsuits, and reputational damage, while those that build trust could gain a competitive edge.
Meta's response—calling the issues isolated and addressing hallucinations—suggests it's trying to contain the fallout. But as AI agents become more common, the pressure on operating systems to limit what apps can see will likely grow. That could mean more friction for users, but also more protection.
For now, the story is a reminder that AI's promise comes with trade-offs. As AI agents start handling more tasks, from shopping to payments, the data they touch becomes more sensitive. Investors should keep an eye on how companies like Meta navigate these privacy challenges, as they could shape the future of AI adoption and regulation.


