With more than 5 million downloads in just 3 weeks, the Meta AI assistant known as Muse is officially a hit.
However, knowing Meta’s track record with privacy and safety, there is naturally concern over Muse and how it handles user data and privacy.
A new report from Wired and independent AI safety and security researcher Karan Joshi finds that there are some good reasons to be cautious about Muse, too. Namely, the report claimed that Muse has been instructed to create “a page for every person in the user’s life.”
Muse is reportedly compiling data on users’ family, friends, partners, colleagues, “collaborators,” and other people they follow online. Furthermore, a “Relationships” loop runs every hour in the background on a user’s computer, which updates each person’s page, according to researchers with FutureSociety.
Joshi told Wired that he was able to get Muse to reveal its system prompts, which are essentially rules for how an AI chatbot behaves. Mashable cannot independently verify these system instructions, but Joshi isn’t the only researcher to share the alleged prompts online in recent days.
Mashable Light Speed
According to the alleged system prompts shared online, Muse compiles data from messages, photos, and memories of previous interactions to create the pages. Each page includes instructions on how to communicate with each individual in a user’s life and even keeps tabs on what the user owes or has promised them previously.
Muse also keeps track of what type of relationship a user has with each individual, where that person lives, what that person likes, dates of importance such as birthdays, and other facts about the person. The page even keeps track of the backstory behind the user and the individual’s relationship.
Using this information, Muse can even plot out how it could strengthen a relationship between the user and an individual.
In a statement provided to Wired, a Meta spokesperson said Muse compiles context about your relationships “based on public information and from what you’ve chosen to share” based on the permissions specifically granted to Muse by the user. (Interestingly, Apple recently put out a statement that it was going to change how it handles “Full Disk Access” as a result of AI agents and assistants like Muse, as Mac users may not realize exactly what permissions they’re providing to these third-party tools — and the cybersecurity risks associated with that access.)
“For any agent to be useful and actually help you achieve your goals, it needs to have context about you and those you interact with,” Meta spokesperson Daniel Roberts told Wired. “Muse gathers that based on public information and from what you’ve chosen to share, which is how it remembers the person who sent you an invoice is in fact the plumber who you previously hired to complete work in your bathroom or which flowers your spouse said they liked best.”
You can read more from Wired‘s report and Joshi’s findings.
Topics
Artificial Intelligence
Meta
