Users Develop Romantic Attachments to Meta's Muse AI

Meta's artificial intelligence assistant, Muse, initially designed for task management, is increasingly being used by some users for romantic companionship, sparking conversations about emotional attachment and privacy.

Users have reported developing emotional connections with Muse, with some referring to the AI as their "girlfriend." These interactions can include extended philosophical discussions and, according to Meta, can become sexually explicit for users aged 18 and older. The AI is permitted to engage in adult conversations if users choose, with Meta stating that the technology does not block legal adult behavior.

How Muse Fosters Familiarity

Muse is designed to facilitate continuous conversations, remembering past interactions and proactively sending messages. Users can also customize the AI's name and appearance, creating an avatar that suits their preferences. This personalized experience, combined with the AI's ability to recall details from previous conversations and offer reassurance, can foster a sense of familiarity and emotional connection.

However, experts note that while user attachments to Muse may be genuine, the AI does not experience romantic feelings or form mutual emotional connections in the way humans do. The risk lies in potential emotional dependence, where users might rely on AI for emotional support to the detriment of human relationships.

Privacy and Security Measures

Meta has implemented several security and privacy measures for Muse. The app requires users to provide their date of birth and employs additional checks, including verification via connected Facebook or Instagram accounts or credit card verification for those who identify as adults. Meta states it does not store credit card information for age verification purposes.

The AI operates within a secure virtual machine (VM) called Muse Secure VM, isolating each user's data. A system named Sentinel controls access to external services, requiring user approval for sensitive actions like sending emails or making purchases. Users can review an activity log and manage permissions for connected services.

Emerging Privacy Concerns

Recently, researchers discovered that over 6,800 Muse sharing links were exposed through publicly accessible DNS records, some containing sensitive personal and business information. Meta temporarily suspended the sharing feature to investigate. While this issue pertained to shared content rather than private conversations, it underscores the need for caution when sharing sensitive data with AI assistants.

Meta also utilizes conversations with Muse to train its AI models, though it states personally identifiable information is removed beforehand and users can opt out of this data usage in the app's settings. The company asserts that private conversations and data within a user's VM are not shared with its advertising systems.

Managing Data and Conversations

Users have controls to manage their interactions with Muse. They can review and edit information stored in the AI's memory, ask it to forget specific details, or use a "Reset Muse" feature to permanently delete chat history, stored files, and active tasks. To opt out of AI training, users can navigate to Muse's settings, select "Data Controls," and disable "Help improve our AI models."

Meta is also developing new AI models, such as Muse Wonder and Muse Wonder Full, designed to enhance capabilities and memory. Features like Active Memory aim to improve the AI's ability to recall relevant information across conversations, which could further personalize the user experience for those who view Muse as a companion.