Is it safe to use Meta Muse?

Meta Muse is an AI that is of great interest at present. Meta Muse brings many interesting things but still brings many risks of revealing user information.

Meta Muse data is accessible

Launched on September 8, Muse is Meta’s response to personal AI assistants. Promoted as “AI helps get work done” and is the solution to help you “Get rid of trivial, time-consuming tasks”, Muse can connect to everything from email to your calendar, to perform tasks on your behalf.

In some aspects, Muse is quite similar to OpenClaw: This is a form of agent AI (agentic AI) that you can chat through the chat frame and directly ask it to complete specific jobs or tasks. However, when experiencing Muse directly, even though this is a “product is quite good”, but it requires access to a lot of sensitive data.

On mobile devices, according to information listed on the Android app store, Muse can collect almost any type of data:

• Photos and videos
• Financial information
• Application information and performance
• Files and documents
• Works within the app
• Personal information: (Name, email address, user ID, address, phone number and other information)
• Contacts
• Browsing activity
• Health and fitness information
• Messages: (Email, SMS/MMS and other in-app messages)
• Audio data
• Location
• Calendar and calendar events

Muse requires very broad permissions, can operate around the clock, interact with your browser, and by default allows data to be used to train AI models (unless you choose to opt out). When combined with the fact that Meta – the creator of Muse – has a notorious reputation for user privacy security, it is not difficult to understand why Muse has raised serious privacy concerns.

So, what happens to the messages and information that a non-Muse user shares with a Muse user (or a Muse virtual assistant)? And is there anything non-Muse users can do to stop Meta from collecting their data?

How does your data appear in Muse – even when you don’t use it?

Although Muse users have actively registered, launched Muse virtual machines (VMs) on Meta’s cloud platform and connected it to their applications and services, people who have contact with them (like friends, colleagues…) do not necessarily agree with that.

Meta says Muse can collect information about people who don’t use the service through a variety of ways, such as when a user connects their Muse assistant to an app or service that contains your data. Users can provide information about themselves and others in the form of text, images, videos or uploaded documents.

Meta for example: “If a Muse user connects their calendar application and asks Muse about scheduled meetings for the week, the names of the participants and event details may appear in the outgoing request and the received response..”

A controversial issue here is what Meta calls “Connectors” (connectors) – act as a bridge between the Muse virtual machine and third-party services through an application programming interface (API). If your Muse account uses the same login information as your Meta Account Center account, Facebook, Instagram, and Threads will automatically be connected.

On launch day, Alexandr Wang, Meta’s AI Director, shared on social network X that Connectors include Gmail, Google Calendar, Outlook, Plaid, OpenTable, Google Docs, Spotify, Function Health, Withings, Tailscale and Peloton. If a service does not have a Connector available, users can ask Muse to make an API call to create a connection to that service on their behalf.

For example, if a Muse user grants permissions and asks their AI to summarize a message thread or draft a response, that AI can read content in messaging apps, email services, or other communication platforms that contain other people’s sensitive information. Travel plans, health concerns, work conversations or financial concerns – all can be handled by an AI assistant (that you don’t give access to and that you don’t use yourself) just because you talk to that Muse user.

For now, it appears that Meta has not taken any significant measures to protect those unwittingly caught up in this process and their data. Muse may scan the web to collect public information relating to you; This is completely consistent with the company’s general approach to data scraping.

Issues with Muse’s privacy policy

Meta says information can be “use or share” as described in Muse’s Privacy Policy. There are two things worth noting here:

Currently, Muse operates on a cloud-based virtual machine (VM). In the future, Meta’s policy states there will be an option to use a secure virtual machine (confidential VM) — an environment that helps “keeps user information private, even to Meta itself.” However, this secure virtual machine environment is not yet available.

Another issue is this: Your interactions — which may include data related to people who don’t use the service — can potentially be used to improve Meta’s AI training.

Is it possible to prevent Muse from accessing your personal data?

It seems like Muse users have more options than those who don’t use the service. Muse users can disable AI training and set access permissions, although the effectiveness of denying certain permissions to the AI ​​remains controversial. For non-Muse users, they only have the right to object based on their local data privacy laws; However, even this permission may not apply to conversations on Muse.

In other words, perhaps users need to carefully consider everything they say to others, even in private emails or messages. If your adversary uses an AI assistant to manage daily tasks, they could unintentionally expose your personal and contact information not only to the AI ​​assistant itself, but also potentially feed this data into the training process of AI models.