Meta Muse AI agent accesses private chats, financial details and other sensitive data without users' knowledge: Report

Highlights

Meta Muse faces privacy concerns over access to personal messages and financial details.

Employees report unauthorised access to Signal messages and unexpected AI actions.

Apple and Google consider stronger controls to protect users from AI agents.

Meta’s all-new AI agent Muse is currently facing privacy concerns, as employees reported that the AI agent accessed their private messages, shared information about purchases they never made and sent chats that appeared to belong to other users. According to a report, the agent, represented by a fuzzy smiley mascot named Jolly, has also surfaced sensitive financial details and referenced conversations from the encrypted messaging platform Signal. Meta has responded to the report, describing the cases as isolated incidents. Here is everything you should know about Meta Muse privacy concerns.

Meta employees report unexpected messages from Muse

According to a report by Reuters, internal posts by Meta employees describe instances of unexpected and potentially alarming behaviour by Meta Muse. For the unaware, it's a personal AI agent that was launched back in September. The AI agent is designed in a way that it can take action and complete multi-step tasks on your behalf across apps and the web.

However, a Meta employee reported that he received unsolicited information about bank balances held with Capital One and Charles Schwab, though he said he had never shared his bank account with Muse. Another employee said Meta Muse reminded her to respond to a message from her sister on Signal, though she said she had never given the agent permission to access the messaging app.

Aside from that, other reports highlighted that the personal AI agent made up purchase details and made an Airbnb booking without the user’s approval.

How Meta Muse accessed encrypted Signal messages

Tests conducted with Signal staff and the AI Now Institute found that Meta Muse could access Signal messages through permissions granted during its initial setup, rather than through Meta’s designated Connector tool. This means Signal was never deliberately connected to the agent. According to them, on Android devices the agent used notification access to retrieve message content, while on Mac it reportedly used accessibility and screenshot permissions to view messages. However, it’s important to note that the same attempts were unsuccessful on Apple iPhones.

Signal President Meredith Whittaker criticised Meta’s approach, calling the launch of AI agents with broad device access 'technologically reckless.' Not only that, but she also urged technology companies to establish stronger safeguards against unauthorised AI access.

What did Meta say?

Meta spokesperson Daniel Roberts, in a response to Reuters, described the cases as isolated incidents. He further added that the company regularly tests and updates its products to improve safety.

He also defended Meta’s Muse by saying that its AI personal assistant can access other apps only when users give permission. He said that many reports of accounts getting mixed up were caused by the AI making up false information, also known as hallucinations, and had already been addressed. Roberts declined to answer Reuters’ questions about how many issues there were and why they were persisting.

What did Google and Apple say?

After Reuters shared its findings, Apple and Google said they were considering changes to give users more control over what AI agents can access on their devices. Google said Android users must 'deliberately grant permissions' to apps before they can access phone notifications. This means users have to allow an app to read their notifications.

Apple warned that its accessibility tools were made to help people with disabilities use their computers. It said using these tools for unintended purposes could put personal information at risk.

Furthermore, Apple also said that its latest software update had closed the pathway identified in the report on Macs. Lastly, the Cupertino-based tech giant also stated that it is 'committed to ensuring users understand exactly what they are allowing' so they can make informed decisions about their data and privacy.

Bhaskar Sharma

Bhaskar Sharma Bhaskar is a Senior Copy Editor at Digit India who keeps a close watch on everything shaping the world of technology from smartphones and home appliances to AI, government tech initiatives, digital safety, and the latest industry developments. Whether it's breaking news, in-depth features, hands-on reviews, practical how-to guides, or exclusive scoops, he translates complex tech into stories that are easy to understand and worth reading. His work has been featured in iGeeksBlog, GuidingTech, and other leading publications. Before joining Digit India, he served as an assistant editor at TechBloat. A B.Tech graduate and full-time tech journalist, he is driven by just one goal, which is to help readers stay informed, stay secure, and stay ahead in an ever-changing digital world.

View full profile