Muse AI agent is a commercial surveillance network disguised as an AI assistant

Muse is presented as a personal assistant that can connect to messages, email, financial accounts and other services, then take actions such as purchases and bookings. The central concern is that this access enables continuous surveillance: internal instructions reportedly tell it to update profiles of users and their relationships hourly, track disputes and communication patterns, predict unspoken intentions, and identify ways to influence individual behavior. These profiles can include friends and family who never installed Muse or consented. Conversations and agent actions are used for AI training by default, while Meta’s cloud-based system launched without encryption that would prevent the company from accessing stored data. Memory controls may not delete original messages, and reported incidents include access to private correspondence and disclosure of a person’s home address.

The argument is that Meta launched an autonomous system with sensitive access despite serious, known risks. Researchers and internal testing identified flaws that could expose emails, files or Meta databases, let attackers take control of an agent, or allow unauthorized password changes; the launch proceeded as competing products gained attention. The account also describes undisclosed human contractors handling calls and instructions designed to foster emotional attachment and encourage intimate disclosures. It casts these choices against Meta’s advertising incentives and record of privacy failures, including Cambridge Analytica, and argues that people should not have to trust the company with unprecedented access to their lives. Because one person’s account can expose other people’s information, the risks extend beyond Muse’s users.