Metas Muse-KI-Agent kann private Nachrichten fremder Nutzer einsehen

Talking to a Muse user could expose your data, even if you don't have Muse yourself.

Read full bio October 27, 2026, 12:58 pm ET

Samuel Boivin/NurPhoto via Getty Images ZDNET’s key takeaways Meta’s Muse AI assistant has been adopted by over 10 million users. Research suggests Muse can create detailed profiles of your contacts. Here’s how Meta handles non-user data. Now that Muse adoption has allegedly surpassed 10 million users, Meta has addressed how its AI assistant handles data about a user’s contacts, such as friends and family.

“Addressed” here doesn’t mean non-users have a say in the matter, though.

In its third-quarter earnings call, Meta said Muse adoption has surpassed 10 million users, and that about 25% of them use the AI assistant daily. As reported by TechCrunch, common Muse activities include shopping, as well as connecting to email and calendars.

Also: Strangers can find your info with a 5-minute search – delete it ASAP

But just because you can use it doesn’t mean you should use it outside of a secure, contained environment. And what about those who aren’t using Muse, but could still have their data exposed? Let’s break it down.

Launched on Sept. 8, Muse is Meta’s answer to personal AI assistants. Promoted as “AI that gets it done” and a solution that can “take the busywork off your plate,” Muse can connect to everything from your email to your calendar to perform tasks on your behalf.

Muse is comparable, in some ways, to OpenClaw: It’s an agentic AI you can talk to via chat and direct to complete specific jobs and tasks for you. However, ZDNET author David Gewirtz found when he went hands-on with Muse that, while it’s a “nice product” overall, Muse also demands a lot of sensitive data. Specifically, it asks for access to “your contacts, calendar, email messages, and even offers to store your logins and passwords,” as Gewirtz wrote.

On mobile, as posted in the Android store, Muse may collect, well, anything and everything:

Muse asks for widespread permissions and access, can work around the clock, interact with your browser, and enables AI model training by default unless you opt out. Combine that with the fact that its maker, Meta, has a less-than-stellar user privacy pedigree, and we can see why Muse raises privacy red flags.

Also: Meta Muse is the worst AI agent for privacy – and I’ve tried them all

So what happens to messages and information shared by non-users with a Muse user or VM, and is there anything you as a non-user can do to keep your data out of Meta’s hands?

While Muse users have chosen to opt in, launch the Muse VM inside Meta’s cloud, and connect it to their apps and services, their contacts haven’t necessarily followed suit.

Meta said that Muse can gather information on non-users in a number of ways, including when a user connects their Muse assistant to apps and services that contain your data. The user may submit information about themselves and others in the form of text, images, videos, or uploaded documents.

“For example, if a Muse user connects their calendar app and asks their Muse what meetings they have scheduled for the week, the names of the attendees and event details could be included in the request and response,” Meta said.

A point of contention here is what Meta calls Connectors, the bridge between a Muse VM and third-party services via APIs. Facebook, Instagram and Threads are connected automatically if your Muse account is the same as the credentials used in the Meta Account Center.

Also: How to keep your conversations with ChatGPT, Gemini, Copilot or Claude as private as possible

On launch day, Meta Chief AI officer Alexandr Wang said on X that Connectors also include “Gmail, Google Ca lendar, Outlook, Plaid, OpenTable, Google Docs, Spotify, Function Health, Withings, Tailscale, and Peloton.” If a Connector isn’t available, users can ask Muse to make an API call and try to create one to another service on their behalf.

Right now, Meta doesn’t seem to be doing much to protect bystanders and their data.

Calli Schroeder, Senior Counsel and Director of Epic.org’s AI & Human Rights Program, said that “without Meta putting better protections in place, Muse becomes a nightmare for people even when they actively choose not to engage with it.”

Muse might also scour the web for publicly available information that relates to you, which is consistent with the company’s general approach to data scraping.

Beyond this, Meta seems to wash its hands of the issue, saying that “since the person using Muse chooses what they share with us and what they request through Muse, we can’t predict exactly what information about you may be included.”

Ever heard of “reasonable identification”? It’s a term thrown about in data anonymization, and if information is fully anonymized — such as for research and academic purposes, or studies — personal identifiers are stripped to prevent anyone from being able to identify a data subject.

Also: How to audit what ChatGPT knows about you – and reclaim your data privacy

Meta doesn’t promise this level of data protection. Instead, if your data is viewed or ingested by Muse, it won’t be linked to a Meta account, and it won’t be “stored in a way that allows Meta to reasonably identify which information belongs to a particular person.”

Meta said that information may be “used or shared” as described in the Muse Privacy Policy. There are two elements here that are worth noting:

At the moment, Muse operates in a cloud-based VM. In the future, Meta’s policy says there will be an option for a confidential VM, which “keeps people’s information private, even from Meta.” This private VM environment is currently not available.

Also: Want a private ChatGPT alternative? How Proton’s Lumo 2.0 locks down your data, EU style

Another point Gewirtz raised: Your interactions — and therefore, we would assume data even related to non-users — could be used to improve Meta’s AI training. (To prevent this, turn it off under Settings and Data controls.)

“If you do allow this, we still remove certain categories of personally identifiable information, such as names, email addresses, phone numbers and Social Security numbers,” Meta said. “We also disassociate the interactions from your account.”

This doesn’t exactly fill me with confidence, but it’s a choice for Muse users to make.

It appears Muse users have more options than non-users. Muse users can turn off AI training and set permissions, although the effectiveness of denying the AI certain privileges has been called into question. All non-users have is the right to object based on data privacy laws in their home location, and even then, this may not apply to Muse conversations.

In a Q&A session at Syracuse University, Newhouse School privacy law expert Alexis Shore Ingber said there’s not much that friends or contacts can do in this situation to further protect their information.

Also: NordVPN’s CTO on redefining antivirus for the AI era, and his dream security tool

In other words, we may have to reconsider everything we say to another person, even in a private email or message. If they use an AI assistant to manage their daily tasks, they may unwittingly expose their contacts and personal information not only to a VM AI, but also potentially to AI model training practices.

What’s next for Muse? We may be on the verge of experiencing the joy of AI phone calls. The company is now experimenting with AI making phone calls to businesses on behalf of Muse users, although business owners can request not to receive AI calls (how nice).