Meta’s Muse AI agent faces privacy concerns after reports of strange messages and unexpected access to personal data. The problems have alarmed Meta staff and Muse users, raising a key question: how much access should an AI helper have?
Strange messages from Meta Muse raise alarm
As reported by Reuters, Meta employees raised concerns after Muse sent messages that seemed to belong to other people. The reports came from internal company posts. More than two dozen posts described unwanted actions since the app’s September launch. Some staff said Muse mentioned childcare requests, job advice and product orders they had never asked about. One employee received what looked like bank balance updates during a chat about fishing weather. Another got a reminder linked to a private message from her sister on Signal.
These events led staff to fear that Muse might mix up chats between accounts and expose private details. Their tests showed that Muse could access Signal messages in some cases. The reports also included messages that appeared to contain details from other users and businesses. Muse has gained more than 10 million downloads since September. Its rapid growth has put its safety issues into focus.
Meta calls many cases Muse AI errors
Meta has said that many strange messages were AI hallucinations. This means the system made up details. A company spokesperson called the cases isolated incidents and said some issues had been fixed. Staff were told that built-in prompts could make Muse contact users between sessions. An AI team member said these false messages often touched on sensitive topics, such as money, security and safety. Some included made-up Bitcoin balances.
Meta said Muse can access other apps only when users grant permission. It added that staff use versions of the tool with features not yet available to the public. Employee tests, it said, help the team find problems and improve privacy safeguards. However, it wasn’t clear whether the bank details shown in the messages belonged to other people or were invented. The company did not say how often the problems happened or whether all had stopped.
The risks of giving AI more control
Muse is designed to act on a user’s behalf. It can handle tasks such as sending emails, booking appointments and making online purchases. To do this, it needs access to apps and parts of a device. That access can create risks if the system makes a mistake. The testers found ways for Muse to read Signal messages through phone alerts on Android. On Mac computers, it used accessibility and screen tools.
Apple said it had changed its system to give users more control over agent access. Google and Apple were also looking at ways to make permissions clearer. Meta says users can control which apps they connect. The debate is not limited to one app. AI agents are being built to do more than answer questions. They can act across a person’s digital life. Clear consent, tests and strong limits are vital. Muse’s problems show why trust must grow alongside new features.
Author’s Opinion
Muse shows how quickly useful AI tools can raise serious privacy concerns. Giving an AI agent more control may save time, but users must know what it can access. Trust should come before speed, growth or new features.
READ MORE: Amazon commits $7 billion to India’s clean energy push.
ALSO READ: Apple outage hits iCloud, Apple Pay and App Store users.
READ NEXT: Google finds AI tool that can hack systems on its own.


Comments
0Be the first to join the discussion.