gizmodo.com
In typical Meta fashion, the company’s popular new AI agent Muse reportedly keeps detailed, regularly updated files on the people in users’ lives.
Nearly a month after its launch, researchers are digging into how Muse works. It turns out the AI agent will openly share internal files and instructions when users ask, offering a peek at how it operates.
One of the somewhat unsettling details is an hourly process that instructs Muse to create and update “a page for every person in the user’s life,” Wired reports.
Independent AI safety and security researcher Karan Joshi extracted a trove of Muse instructions and system prompts through the app’s regular chat interface and shared the findings with Wired. Separately, The Future Society published system instructions from an exported copy of a Muse virtual machine.
“What it seemed like to me—from all these prompts, system skills data, and things that they’re feeding into Muse—is that they want to understand your relationships that you have with real people,” Joshi told Wired. “They’re trying to know you like a friend, which is honestly pretty creepy.”
The new details come just weeks after Meta launched Muse on Sept. 8 as a personal AI agent that can handle tasks like sending emails, booking travel, shopping, and tracking goals. To do all that, users give Muse access to a wide range of their apps and accounts. Muse has since climbed to the number-one spot on Apple’s U.S. App Store chart for free iPhone apps.
But privacy and security concerns have emerged almost immediately.
For instance, Amazon blocked Muse from shopping on its site in September, saying Meta had not notified the company in advance that the agent would access its online shop. The company also raised concerns about how Muse handles customer credentials and account data.
Meta also recently patched a zero-day vulnerability in the Mac version of Muse that could allow malware already running on a user’s computer to hijack the agent and take advantage of permissions the user had previously granted it.
Now, we have a better idea of just how much Muse keeps track of about users and the people in their lives, even if those people don’t use Muse themselves.
According to system instructions published by The Future Society, these relationship pages can include a person’s name, how they are connected to the Muse user, recurring topics such as an upcoming apartment move or savings goal, and important dates like birthdays and anniversaries.
Muse can also keep a history of the relationship, including events or milestones, as well as information about how close the user is to the person, what their relationship is built on, how they interact, and what the relationship appears to need at the moment.
“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,” a Meta spokesperson told Gizmodo in an emailed statement. “Muse gathers that based on public information and from what you’ve chosen to share, which is how it remembers the person who just sent you an invoice is in fact the plumber who you previously hired to complete some work in your bathroom or which flowers your spouse said they liked best.”
The spokesperson added that the information is stored in each user’s dedicated secure virtual machine, which Meta says cannot be accessed by other users’ agents.
