Goal has gotten millions of people to download Muse in just a month, but its new artificial intelligence agent is raising serious doubts about the privacy it offers. A Wired investigation has discovered that the system creates detailed profiles of its users’ friends, family and co-workers without those affected being aware of it.
Muse is an AI agent that Meta launched on September 8 in the United States. Unlike conventional chatbots, it does not limit itself to answering questions, but You can perform tasks on your own, such as sending emails, organizing trips, shopping, or managing appointments.
The service, which uses Meta’s Muse Spark model, works through a mobile app, website and WhatsApp. It allows you to connect email accounts, calendars and other personal services to automate daily activities and also can continue working when the user closes the application. For now it is not available in Spain, with no scheduled date for its arrival.
The problem discovered by Wired has its origins in Muse’s internal instructions, which several researchers have managed to extract through conversations with the agent himself. one of them is Karan Joshispecialized in artificial intelligence security, who shared his findings with the media.
Among those instructions is the creation of a ‘page for every person in the user’s life’a procedure designed to be executed every hour. The documents indicate that these profiles can include information about where someone lives, what they do, their interests, birthdays, shared events and the relationship they have with the user.
Files can start with a few pieces of information and Expand as Muse learns more about conversations and connected services. They even include sections dedicated to the emotional closeness between people and suggestions to improve your relationships, although it is not proven that all profiles contain this information.
‘The impression I got, from all those instructions, the data on the system’s capabilities and the information they are providing to Muse, is that They want to understand the relationships you have with real people‘ says Joshi. The researcher considers that ‘they try to get to know you like a friend would, which is honestly quite disturbing’.
Meta told Wired that this information allows the assistant to remember useful details and offer a personalized experiencelike knowing the favorite flowers of the user’s partner. The company ensures that the data comes from public information or what people decide to share, and that users can delete memories or disconnect services.
Privacy concerns come after 404 Media revealed that Meta engineers detected several serious vulnerabilities shortly before launch. At least one would have allowed someone to manipulate Muse to bypass security barriers that limit their access to Meta systems and enter the company’s internal databases without authorization.
The problems were so important that they came to the attention of Mark Zuckerberg and several teams worked on a ‘race against time’ to solve them before the premiere. An internal source claimed that protections had been hastily applied and warned that ‘many senior engineers believe it is inevitable that we will suffer a massive data breach as a result of Muse’.
Meta assures that it has reinforced the security of the agent and that each user has an independent virtual machine to protect their information. However, the Wired revelations raise another problem that goes beyond technical vulnerabilities, as a person can end up being part of an AI’s records without ever having used the service.