Meta assistant reportedly builds unrequested profiles of users' contacts

Researchers recently discovered that Meta's AI assistant, named Muse, is programmed to automatically generate detailed profiles for people mentioned by users, according to a report from WIRED. Security researcher Karan Joshi extracted the system's underlying instructions through standard chat prompts. The code suggests the software builds individual entries for friends, family, and acquaintances without requiring those people to sign up or give permission.

This practice expands privacy concerns beyond just the person interacting with the chatbot. If someone talks to an AI tool about a relative's health issue or a friend's new job, that information could end up stored in a permanent profile about someone who never agreed to use the service. Most people expect their casual conversations about loved ones to remain private, but automated tracking turns shared personal details into structured dossiers without clear consent.

It remains uncertain how Meta plans to handle these files long term, or whether individuals will ever have the right to inspect or delete profiles made about them covertly. Meta maintained that it intended for these system instructions to be accessible in the interest of transparency, but the boundary between helpful context for a virtual assistant and non-consensual data harvesting remains blurry. This raises an urgent question about whether tech firms will face new regulatory pushback for building shadow databases of non-users through AI conversations.